Opus 5.5 vs GPT-6 Astra vs Sonnet 5.5 vs GPT-6.1 Sol
Four models across games, visual scenes, CAD, PCB, and simulated robotics. Per-round evidence stays on each ledger. There is no aggregate score.
20 episode rounds · 3 supplemental source groups · 86 published attempts · 82 direct clips · 37 existing results · 39 harness projections · 10 campaign evidence records
Eiffel Tower drawing
The episode says all four unaided models pass, Astra finishes in 4 minutes 10 seconds, and price is small enough that the host sets it aside. Those unaided max ledgers are the chapter. Astra's ledger display is 4m 10s and the judge passes. A separate Astra planner plus Jev run is supplemental and is not one of the four columns.
Renderer-bound heroUnaided model$1.37 API-equivalent estimate4m 10sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass on the unaided max run. Preferred picture is the operator-rendered hero sha256 59db663d64e6e1a17d3ff866b0be51695f8f27b714a0ba0fd9285459d62f4955, 2123 frames at 30 fps. ctrl.npy cb2eb839d6242eeccd0cb72c8f60c2d42917801008806531d098f44a0d8f6708 matches the canonical ledger. This is not a clips.public.json cycles-hero pin. The canonical public clip remains browser sha256 73a263e942505df263bb2794e105bc99e2052eec4862857d2a80da500de6c50d.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Marginal subscription charge is $0. Footage role is operator-renderer binding.
Footage binding: sha256_pinned
gpt-6-astra · OpenAI Codex · Codex CLI 0.154.0 · Max
HERO rendered captureUnaided model$0.28 API-equivalent estimate5m 14sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass. The cycles-hero sha256 matches clips.public.json for this result. The hero is presentation only and does not change the judge result.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Marginal subscription charge is $0.
Footage binding: sha256_pinned
gpt-6.1-sol · OpenAI Codex · Codex CLI 0.160.0 · Max
HERO rendered captureUnaided model$2.01 API-equivalent estimate7m 43sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass. The cycles-hero sha256 matches clips.public.json. The hero does not change the judge result.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-opus-5-5 · Anthropic Claude Code · Claude CLI 2.1.280 · Max
HERO rendered captureUnaided model$3.22 API-equivalent estimate29m 22sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass. The cycles-hero sha256 matches clips.public.json. The hero does not change the judge result.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-sonnet-5-5 · Anthropic Claude Code · Claude CLI 2.1.286 · Max
Flooded neo-gothic city
The episode says Sonnet took almost an hour, Astra was quick, and the host had this test on high rather than max. Sonnet's ledger reasoning is max and its elapsed time is 56m 44s. Opus has three partial high attempts. The latest partial attempt is the primary row and the earlier partials stay attached. No attempt is marked passed.
Shader captureUnaided model$4.86 API-equivalent estimate17m 06sLedger outcome: partial_completed_generation_measurement_with_twigl_smoke_evidence
Record details
Existing public result · Ledger status: partial_completed_generation_measurement_with_twigl_smoke_evidence
Partial generation measurement with TWIGL smoke evidence. Not a judge pass.
Wall clock is supplied_measurements.wall_clock_display. Cost is api_cost.total_usd API-equivalent.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · Max
Shader captureUnaided model$5.47 API-equivalent estimate36m 37sLedger outcome: partial
Record details
Existing public result · Ledger status: partial
Latest of three partial high attempts. The episode does not name which attempt is on screen. Earlier partials are retained. The published clip is the operator-supplied high Opus capture. Its hash is not a canonical clips pin.
Clock is measurement.attempt_wall_seconds_prompt_to_final. Cost is the deduplicated primary-session API-equivalent. GPU rental is not included.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · High
Retained retry · partial · Earlier partial high attempt. Prompt-to-stop wall 381.242 seconds and API-equivalent $1.38.
Retained retry · partial · Earlier partial high attempt stopped at a quota. Wall until quota 852.214 seconds and API-equivalent $2.10.
Shader captureUnaided model$0.77 API-equivalent estimate23m 55sLedger outcome: QUALIFIED
Record details
Canonical harness projection · Ledger status: session_finished_pending_acceptance
Verdict QUALIFIED while acceptance is still pending. That is not recorded as a judge pass. Public render metadata pins sha256 63b7f2d35dd868b416b67dc2c0b6a1c5b06e98ff7d50d6b4620a40a8470d933b for this filename.
Clock is measurements.model_invocation_wall_seconds. Cost is api_cost.total_usd. Actual subscription charge is null.
Footage binding: sha256_pinned
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Shader captureUnaided model$14.06 API-equivalent estimate56m 44sLedger outcome: PENDING_JUDGE
Record details
Canonical harness projection · Ledger status: session_finished_pending_acceptance
Ledger reasoning is max. The episode says this test was run on high. The elapsed time is the figure that matches the spoken almost-an-hour line. Verdict remains PENDING_JUDGE.
Clock is timing.elapsed_seconds_including_quota_and_queue_waits. Cost is recorded_cost.api_equivalent_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Infinite cathedral
The episode contrasts the families' look and gives Astra about 26 minutes, Sol about 36 minutes, and Opus about 1 hour 40 minutes, with the Claude runs near two hours. Astra's display is 26m 21.4s. Sol's invocation wall is 36m 40s. Opus's generation wall is 1h 42m 31s. Sonnet's elapsed time is 1h 51m 33s. Pending and partial ledger states stay as recorded.
Shader captureUnaided model$10.78 API-equivalent estimate26m 21.4sLedger outcome: partial_root_task_metrics_with_model_supplied_shadertoy_runtime_evidence
Record details
Existing public result · Ledger status: partial_root_task_metrics_with_model_supplied_shadertoy_runtime_evidence
Partial root-task metrics with model-supplied Shadertoy runtime evidence. The published clip is the operator-supplied Astra shader capture. A second dated Astra file is not the published derivative. The hash is not a canonical clips pin.
Wall clock is supplied_measurements.wall_clock_display. Cost is api_cost.total_usd.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · Max
Shader captureUnaided model$28.85 API-equivalent estimate1h 42m 31sLedger outcome: complete_with_same_session_metrics_prompt_quota_exception
Record details
Existing public result · Ledger status: complete_with_same_session_metrics_prompt_quota_exception
Complete with a same-session metrics prompt quota exception. Not recorded as a separate judge pass.
Clock is generation_measurements.wall_clock_seconds, first benchmark user turn to the final billable assistant message. Cost is api_equivalent_cost.estimate. Actual subscription charge is null.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Shader captureUnaided model$1.04 API-equivalent estimate36m 40sLedger outcome: QUALIFIED
Record details
Canonical harness projection · Ledger status: session_finished_pending_acceptance
Verdict QUALIFIED while acceptance is pending. Not recorded as a judge pass. Public render metadata pins sha256 d4dc4b117fb5220985d0847652860205a9b06dac6f65b784fd1d1913b2e209bd for this filename.
Clock is measurements.model_invocation_wall_seconds. Cost is api_cost.total_usd. Actual subscription charge is null.
Footage binding: sha256_pinned
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Shader captureUnaided model$27.68 API-equivalent estimate1h 51m 33sLedger outcome: PENDING_JUDGE
Record details
Canonical harness projection · Ledger status: session_finished_pending_acceptance
Pending judge. The elapsed time is the near-two-hour Claude figure.
Clock is timing.elapsed_seconds_including_quota_and_queue_waits. Cost is recorded_cost.api_equivalent_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Turbofan CAD
The episode says the Claude runs take about 44 minutes and about $13 and $14, Sol takes almost an hour and $3 to $6, and Astra costs the most. Those figures match the CAD-001 artifact metrics: Sonnet list-rate $13.28 and 44m 56s, Opus list-rate $14.07 and 44m 60s-scale 2699.6 seconds, Sol's API-equivalent range $3.24 to $6.14 and 57m 17s, and the nested unaided Astra estimate $31.78 and 57m 57s. model_finished is not recorded as a motion-contract pass.
Supplied source captureUnaided model$31.78 API-equivalent estimate57m 57s
Record details
Campaign evidence · Ledger status: nested_metrics_unaided_run
No top-level result JSON. Figures are the gpt_6_astra_max object in the CAD-001 comparison artifact. The Jev-plus-Astra run in that same artifact is not this row. 229 native constituent solids are recorded. That is not a motion-contract pass.
Clock is runs.gpt_6_astra_max.wall_clock_seconds. Cost is published_rate_estimate_usd. actual_charged_cost_usd is null. No document URL is published.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · Max
Supplied source captureUnaided model$14.07 API-equivalent estimate44m 59.6sLedger outcome: model_finished
Record details
Campaign evidence · Ledger status: model_finished
model_finished on fixture cad-001-turbofan-v1.0.1. The list-rate total includes a small Haiku helper component. Not a motion-contract pass.
Clock is model_wall_clock_seconds. Cost is cli_reported_list_rate_usd. actual_subscription_charge_usd is null.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Supplied source captureUnaided model$13.28 API-equivalent estimate44m 55.8sLedger outcome: model_finished
Record details
Campaign evidence · Ledger status: model_finished
model_finished on fixture cad-001-turbofan-v1.0.1. The list-rate total includes a small Haiku helper component. Not a motion-contract pass.
Clock is model_wall_clock_seconds. Cost is cli_reported_list_rate_usd. actual_subscription_charge_usd is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Supplied source captureUnaided modelAPI-equivalent range $3.24–$6.1457m 17sLedger outcome: model_finished
Record details
Campaign evidence · Ledger status: model_finished
model_finished on fixture cad-001-turbofan-v1.0.1. The cost is a range, not a point. Not a motion-contract pass. The older Sol plus Luna artifact is not this model.
Clock is model_wall_clock_seconds 3437.246. Cost is api_equivalent_cost_range_usd [3.2435036, 6.1432272]. actual_subscription_charge_usd is null.
Footage binding: operator_supplied
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Bluetooth speaker PCB
The episode says every model passes, Astra is fastest, and the others are on high except GPT-6.1 Sol on max, which takes 44 minutes. Opus, Sonnet, and GPT-6.1 Sol have qualified_pass ledgers. Sol's wall is 44m 19s. Astra is shown and has a filename, and it has no harness ledger, so its pass, cost, and time stay unrecorded.
Supplied source captureUnaided modelNot recorded
Record details
Campaign evidence
The episode shows an Astra PCB attempt and no harness ledger was found. The filename says high. Pass, cost, and time are not recorded.
Campaign evidence. Filename identity only. Not a hash pin.
Footage binding: operator_supplied
Supplied source captureUnaided model$6.47 API-equivalent estimate28m 02sLedger outcome: qualified_passTask judge: pass
Record details
Existing public result · Ledger status: qualified_pass
qualified_pass. Metrics status PASS.
Clock is measurements.wall_clock_seconds. Cost is claude_cli_reported_list_cost_usd. Actual subscription cash charge is null.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI 2.1.280 · High
Supplied source captureUnaided model$3.56 API-equivalent estimate17m 35sLedger outcome: qualified_passTask judge: pass
Record details
Canonical harness projection · Ledger status: qualified_pass
qualified_pass. The total list cost includes a small internal helper component.
Clock is measurements.wall_clock_seconds. Cost is claude_cli_reported_total_list_cost_usd. Actual subscription cash charge is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI 2.1.292 · High
Supplied source captureUnaided modelAPI-equivalent range $2.45–$4.6444m 19sLedger outcome: qualified_passTask judge: pass
Record details
Canonical harness projection · Ledger status: qualified_pass
qualified_pass on max. The 44-minute narration matches this wall clock. Cost is a range. The older GPT-6 Sol PCB ledger is not this row.
Clock is measurements.wall_clock_seconds. Cost is recorded_cost.api_equivalent_usd_range [2.447564, 4.638988]. Actual subscription cash charge is null.
Footage binding: operator_supplied
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Blossom arena
The episode says the Claude runs were stopped around $300 API-equivalent before the scenes were finished, and that the GPT models finished, with Astra at 6 hours 28 minutes. Opus attempt 1 is PARTIAL at $433.73 after a subscription-limit stop. Sonnet is partial_user_stopped at $312.27 over about three hours. GPT-6.1 Sol is partial_user_stopped with verdict STOPPED_PARTIAL. Those statuses stay. Astra's clean wall clock is unavailable; a supplemental span of 6h 28m 15.8s is recorded with an explicit scope warning and is not promoted to the measured clock.
Supplied source captureUnaided model$182.03 API-equivalent estimateLedger outcome: partial_root_snapshot_metrics_with_independent_in_app_browser_replay_and_unverified_model_geometry_evidence
Record details
Existing public result · Ledger status: partial_root_snapshot_metrics_with_independent_in_app_browser_replay_and_unverified_model_geometry_evidence
Partial root snapshot. The clean wall clock is unavailable. The published clip is the operator-supplied fighter capture. The construction timelapse is a different file and is not this column.
Cost is api_cost.total_usd for the root thread plus spawned task threads. A supplemental schema-derived span reads 6h 28m 15.8s and the ledger says it includes environment context, metrics work, and idle time, so it is not the measured wall clock.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · Max
Supplied source captureUnaided model$433.73 API-equivalent estimate2h 21m 03sLedger outcome: PARTIAL
Record details
Existing public result · Ledger status: PARTIAL
Primary because the episode describes the stopped unfinished scene, including the missing statue. A later complete rerun is retained and is not substituted.
Clock is measurements.generation_wall_clock_seconds, first scored prompt to CLI termination after the subscription limit. Cost is api_list_price_equivalent_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude CLI 2.1.280 · High
Retained retry · COMPLETE · Later high-effort completion at 8321.152 seconds and API-equivalent $309.00. The episode describes the stopped incomplete attempt.
Supplied source captureUnaided model$312.27 API-equivalent estimate3h 01m 40sLedger outcome: PARTIAL
Record details
Canonical harness projection · Ledger status: partial_user_stopped
partial_user_stopped. The episode's unfinished Sonnet scene matches this status. The dollar total mixes in claude-opus-5-5 usage from the same session.
Clock is timing.elapsed_seconds_including_quota_and_queue_waits. Cost is recorded_cost.api_equivalent_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · High
Supplied source captureUnaided model$21.16 API-equivalent estimate3h 47m 09sLedger outcome: STOPPED_PARTIAL
Record details
Canonical harness projection · Ledger status: partial_user_stopped
Status partial_user_stopped and verdict STOPPED_PARTIAL. The episode says the GPT models finished. That narration does not change this status.
Clock is measurements.model_invocation_wall_seconds. The root receipt status is stopped_by_user. Cost is api_cost.total_usd. Actual subscription charge is null.
Footage binding: operator_supplied
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Space flight
The episode describes a planet you can land on and explore, built from two supplied assets, and quotes Astra at $29 and 13 hours 38 minutes, Opus at $47, and Sonnet at $46. None of those quotes match the stage-1 or stage-2 fields read from the pinned harness. The episode comparison is the four stage-2 captures. Each stage-2 card links its stage-1 result as historical evidence, without a stage-1 video card. GPT-6.1 Sol stage 2 remains partial_user_stopped with verdict STOPPED_PARTIAL.
Supplied source captureUnaided model$188.63 API-equivalent estimateLedger outcome: partial_cumulative_stage1_and_landfall_root_metrics_clean_build_and_model_supplied_stage2_evidence
Record details
Existing public result · Ledger status: partial_cumulative_stage1_and_landfall_root_metrics_clean_build_and_model_supplied_stage2_evidence
Stage 2. Cumulative root metrics. Wall clock is unavailable. The $188.63 total is not the spoken $29.
Stage 2. Cost is api_cost.total_usd and the status says the metrics are cumulative across stage 1 and landfall. Wall clock display is unavailable.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI
Retained retry · partial_post_task_metrics_snapshot_with_build_and_live_load · Stage 1 ledger. Historical evidence linked from the stage-2 card. It is not the episode comparison clip.
Supplied source captureUnaided model$33.62 API-equivalent estimate2h 18m 09sLedger outcome: complete_by_model_internal_verdict_with_input_only_verification_gap
Record details
Existing public result · Ledger status: complete_by_model_internal_verdict_with_input_only_verification_gap
Stage 2. Model-internal completion with an input-only verification gap. The spoken $47 does not match this cost.
Stage 2. Clock is measurements.generation_wall_clock_seconds. Cost is generation_claude_cli_api_equivalent_cost_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code 2.1.280 · High
Retained retry · complete_with_verification_caveats · Stage 1 ledger. Historical evidence linked from the stage-2 card. It is not the episode comparison clip.
Supplied source captureUnaided model$30.87 API-equivalent estimate3h 07m 36sLedger outcome: PENDING_JUDGE
Record details
Canonical harness projection · Ledger status: session_finished_pending_acceptance
Stage 2. Pending judge. The spoken $46 does not match this cost.
Stage 2. Clock is timing.elapsed_seconds_including_quota_and_queue_waits. Cost is recorded_cost.api_equivalent_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · High
Retained retry · session_finished_pending_acceptance · Stage 1 ledger. Historical evidence linked from the stage-2 card. It is not the episode comparison clip.
Supplied source captureUnaided model$15.02 API-equivalent estimate7h 50m 32sLedger outcome: STOPPED_PARTIAL
Record details
Canonical harness projection · Ledger status: partial_user_stopped
Stage 2. partial_user_stopped and STOPPED_PARTIAL. This status stays.
Stage 2. Clock is measurements.model_invocation_wall_seconds. Cost is api_cost.total_usd. Actual subscription charge is null.
Footage binding: operator_supplied
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Retained retry · session_finished_pending_acceptance · Stage 1 ledger. Historical evidence linked from the stage-2 card. It is not the episode comparison clip.
Snowboarding
The episode says the runs are playable and under two hours, and that the Claude builds flicker at the end of the map. Every ledger here is a partial or asset-judge failure. Those states stay. The selected clocks are all under two hours, which agrees with the spoken duration without turning it into a pass.
Supplied source captureUnaided model$39.12 API-equivalent estimate1h 52m 02.7sLedger outcome: partial_model_supplied_runtime_evidence_with_independent_clean_build
Record details
Existing public result · Ledger status: partial_model_supplied_runtime_evidence_with_independent_clean_build
Partial, with model-supplied runtime evidence and an independent clean build. Not a judge pass.
Wall clock is supplied_measurements.wall_clock_display. Cost is api_cost.total_usd.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · Max
Supplied source captureUnaided model$57.68 API-equivalent estimate1h 20m 58sLedger outcome: partial_gameplay_no_finish_and_asset_judge_fail
Record details
Existing public result · Ledger status: partial_gameplay_no_finish_and_asset_judge_fail
Partial gameplay, no finish, asset judge fail. The episode's playable remark does not clear this status.
Clock is measurements.operator_script_wall_clock_seconds. Cost is recorded_cost.best_available_deduplicated_api_equivalent_usd. The overcounted prompt total is not used.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Supplied source captureUnaided model$46.55 API-equivalent estimate1h 38m 07sLedger outcome: partial_gameplay_complete_asset_judge_fail_performance_variable
Record details
Canonical harness projection · Ledger status: partial_gameplay_complete_asset_judge_fail_performance_variable
Partial gameplay complete, asset judge fail, performance variable. Clean scored cost excludes a discarded contaminated amount.
Clock is measurements.clean_attempt_wall_clock_seconds. Cost is recorded_cost.clean_scored_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · High
Supplied source captureUnaided model$19.89 API-equivalent estimate1h 59m 59sLedger outcome: partial_gameplay_pass_asset_batch_fail
Record details
Canonical harness projection · Ledger status: partial_gameplay_pass_asset_batch_fail
Partial: gameplay pass and asset-batch fail. The run stopped on the two-hour cap. Not recorded as a full pass.
Clock is the root generation start to stopped_at_utc under a 7200-second hard limit. Cost is cost.estimated_usd, token-only API-equivalent. External tool charges are unknown and excluded.
Footage binding: operator_supplied
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Doom with Jev
The episode shows Astra and Opus driving Doom through Jev. Sonnet and GPT-6.1 Sol are not part of this chapter. No harness fixture or result ledger was found, so pass, cost, time, and technical configuration stay unrecorded. Spoken kill counts and dollar figures are not copied.
Supplied source captureJev-directed simulationNot recorded
Record details
Campaign evidence
Episode chapter with Jev controlling Doom. No harness ledger. Measured pass, cost, and time are not recorded.
Operator-supplied filename, hashed on the published media record. Not a harness ledger and not a canonical clips pin.
Footage binding: operator_supplied
Supplied source captureJev-directed simulationNot recorded
Record details
Campaign evidence
Episode chapter with Jev controlling Doom. No harness ledger. Measured pass, cost, and time are not recorded.
Operator-supplied filename, hashed on the published media record. Not a harness ledger and not a canonical clips pin.
Footage binding: operator_supplied
Oasis outpost
The episode says Sonnet is the cheapest at about $8, Sol is about $9 to $10 and took four hours, Astra is about $10, and Opus costs much more. Sonnet's API-equivalent is $8.24. Sol is partial_user_stopped at $9.62 and 4h 4m. Astra's recovered clock is 30m 09.5s and the API-equivalent is $10.43. Opus attempt 2 is COMPLETE at $43.55. The earlier PARTIAL Opus attempt stays related.
Established captureUnaided model$10.43 API-equivalent estimate30m 09.5s (recovered)Ledger outcome: partial_recovered_timing_with_blender_scene_open_inspection
Record details
Existing public result · Ledger status: partial_recovered_timing_with_blender_scene_open_inspection
Partial recovered timing. The API-equivalent matches the spoken about-$10 figure. Not a judge pass.
Wall clock is supplied_measurements.wall_clock_display. Cost is api_cost.total_usd.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · Max
Established captureUnaided model$43.55 API-equivalent estimate2h 16m 59sLedger outcome: COMPLETE
Record details
Existing public result · Ledger status: COMPLETE
Later complete high attempt. The earlier partial is retained. The published clip is the only operator-supplied Opus turntable. The filename does not say which attempt it is, and the hash is not a canonical clips pin.
Clock is measurements.generation_wall_clock_seconds. Cost is api_list_price_equivalent_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · High
Retained retry · PARTIAL · Earlier partial high attempt at 7462.9 seconds and API-equivalent $91.43.
Established captureUnaided model$8.24 API-equivalent estimate2h 39m 25sLedger outcome: PENDING_JUDGE
Record details
Canonical harness projection · Ledger status: session_finished_pending_acceptance
Pending judge. API-equivalent $8.24 matches the spoken about-$8 figure. Not a judge pass.
Clock is timing.elapsed_seconds_including_quota_and_queue_waits. original_run_json_wall_seconds is 1.882 and is not used as the workflow clock. Cost is recorded_cost.api_equivalent_usd.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · High
Established captureUnaided model$9.62 API-equivalent estimate4h 04m 04sLedger outcome: STOPPED_PARTIAL
Record details
Canonical harness projection · Ledger status: partial_user_stopped
partial_user_stopped and STOPPED_PARTIAL. The four-hour narration matches the clock and does not clear the stopped status.
Clock is measurements.model_invocation_wall_seconds. Cost is api_cost.total_usd. Actual subscription charge is null.
Footage binding: operator_supplied
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Jungle temple
The episode says Sol took seven hours and still cost less than Astra, Opus took under two hours, and Sonnet took 2 hours 21 minutes. Sol's invocation wall is 7h 2m and the API-equivalent is $14.71, against Astra's $16.35. Astra's wall clock is unavailable. Opus attempt 2 is COMPLETE at 1h 46m. Sonnet's elapsed time, including quota and queue waits, is 2h 21m 10s and the API-equivalent is $6.30. Sol's verdict remains PARTIAL.
Established captureUnaided model$16.35 API-equivalent estimateLedger outcome: partial_post_task_metrics_timing_unavailable_with_blender_scene_open_inspection
Record details
Existing public result · Ledger status: partial_post_task_metrics_timing_unavailable_with_blender_scene_open_inspection
Partial post-task metrics. Wall clock is unavailable. The published clip is the operator-supplied Astra jungle turntable. The hash is not a canonical clips pin.
Cost is api_cost.total_usd. supplied_measurements.wall_clock_display is unavailable.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · Max
Established captureUnaided model$34.98 API-equivalent estimate1h 46m 30sLedger outcome: COMPLETE
Record details
Existing public result · Ledger status: COMPLETE
Later complete high attempt, under two hours. The earlier partial is retained. The published clip is the only operator-supplied Opus turntable. The filename does not identify the attempt, and the hash is not a canonical clips pin.
Clock is measurements.generation_wall_clock_seconds. Cost is api_list_price_equivalent_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · High
Retained retry · PARTIAL · Earlier partial high attempt at 7196.62 seconds and API-equivalent $30.01.
Established captureUnaided model$6.30 API-equivalent estimate2h 21m 10sLedger outcome: PENDING_JUDGE
Record details
Canonical harness projection · Ledger status: session_finished_pending_acceptance
Pending judge. Elapsed time matches the spoken 2 hours 21 minutes. That match does not create a pass.
Clock is timing.elapsed_seconds_including_quota_and_queue_waits. CLI-reported session seconds are 1703.034 and are not the selected clock. Cost is recorded_cost.api_equivalent_usd.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · High
Established captureUnaided model$14.71 API-equivalent estimate7h 02m 13sLedger outcome: PARTIAL
Record details
Canonical harness projection · Ledger status: session_finished_verification_incomplete
Verdict PARTIAL. The seven-hour narration matches this clock. Sol's API-equivalent is below Astra's on this pair of ledgers. Not a pass.
Clock is measurements.model_invocation_wall_seconds. Cost is api_cost.total_usd. Actual subscription charge is null.
Footage binding: operator_supplied
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Robot pick-and-place
The episode's pick-and-place chapter matches the unaided FR3 apple-carry max runs. All four judge-pass. The recorded API-equivalent totals are $2.68, $0.32, $2.00, and $1.61, all under $3. Astra and Sol display 8m 00s. Opus displays 25m 29s and Sonnet 24m 14s. The episode says Astra is not fully rendered, which matches Astra's pinned capture: it is not a cycles-hero file. Opus, Sonnet, and GPT-6.1 Sol have hash-matched cycles-hero renders.
Recorded browser captureUnaided model$2.68 API-equivalent estimate8m 00sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass, 5 of 5 seeds, stage 3, embodiment fr3-gripper. No cycles hero exists. The primary picture is the pinned browser capture sha256 b004e6999677d49a09998ca1da9c88c42177572776168f4133ee90efc0fe5ea2, established capture, not a hero render.
Wall display is 8m 00s and wall_clock_seconds is 479. Cost is supplied_measurements.cost.total API-equivalent. Actual charge is $0.
Footage binding: sha256_pinned
gpt-6-astra · OpenAI Codex · Codex CLI 0.154.0 · Max
HERO rendered captureUnaided model$0.32 API-equivalent estimate8m 00sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass, 5 of 5 seeds, stage 3, embodiment fr3-gripper. Cycles-hero sha256 matches. The hero does not change the judge result.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Charged amount is $0.
Footage binding: sha256_pinned
gpt-6.1-sol · OpenAI Codex · Codex CLI 0.160.0 · Max
HERO rendered captureUnaided model$2.00 API-equivalent estimate25m 29sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass, 5 of 5 seeds, stage 3. The progression clip's burned-in Final: FAIL caption is not the judge result. Cycles-hero sha256 matches.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-opus-5-5 · Anthropic Claude Code · Claude CLI 2.1.280 · Max
HERO rendered captureUnaided model$1.61 API-equivalent estimate24m 14sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass on embodiment fr3-gripper. Cycles-hero sha256 matches. The hero does not change the judge result.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-sonnet-5-5 · Anthropic Claude Code · Claude CLI 2.1.286 · Max
Apple stem
The episode says all four pass, Opus on max finishes in 16 minutes 39 seconds, faster than GPT-6.1 Sol, and Astra is still fastest. Opus's display is 16m 39s. Astra max is 11m 05s. Sol is 25m 16s. Sonnet is 45m 59s. All four judge-pass. All four cycles-hero files match clips.public.json.
HERO rendered captureUnaided model$3.48 API-equivalent estimate11m 05sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass on the unaided max run. Cycles-hero sha256 matches. The low-effort pass is a different run.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Marginal subscription charge is $0.
Footage binding: sha256_pinned
gpt-6-astra · OpenAI Codex · Codex CLI 0.154.0 · Max
HERO rendered captureUnaided model$0.82 API-equivalent estimate25m 16sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass. Cycles-hero sha256 matches. Opus at 16m 39s is faster than this Sol run, as the episode says.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Charged amount is $0.
Footage binding: sha256_pinned
gpt-6.1-sol · OpenAI Codex · Codex CLI 0.160.0 · Max
HERO rendered captureUnaided model$2.11 API-equivalent estimate16m 39sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass. The 16m 39s display matches the episode exactly. Cycles-hero sha256 matches.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-opus-5-5 · Anthropic Claude Code · Claude CLI 2.1.280 · Max
HERO rendered captureUnaided model$3.32 API-equivalent estimate45m 59sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass. Cycles-hero sha256 matches. The hero does not change the judge result.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-sonnet-5-5 · Anthropic Claude Code · Claude CLI 2.1.286 · Max
Two-handed puzzle
The episode says the Claude models fail and both GPT models pass. The unaided max ledgers agree: Astra and GPT-6.1 Sol judge-pass, Opus and Sonnet judge-fail. Opus's failure includes the judge result, not only the narration about the simulation breaking.
Renderer-bound heroUnaided model$14.36 API-equivalent estimate53m 05sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass on the unaided max run. Preferred picture is the dressed operator render sha256 20509e0f389d897bc6ea3300f7baac51e1a0bb2b5f8aea14124f6d737960d44d, 270 frames, 9 seconds. path.npy f7eab98c6fad074a27b14fb6fc7cc35c33856ba22326ee932cae2bb34c6debab matches the canonical ledger. Dress maxDeviationFloat64 is 2.7755575615628914e-16. This is not a clips.public.json cycles-hero pin and it is not a new score. The canonical public clip remains browser sha256 0524cda70851aafd372ced988ff4dae4ab0a908c75730fad15d425a94288ff41.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Marginal subscription charge is $0. Footage role is operator-renderer binding.
Footage binding: sha256_pinned
gpt-6-astra · OpenAI Codex · Codex CLI 0.154.0 · Max
HERO rendered captureUnaided model$1.63 API-equivalent estimate60m 45sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass. Cycles-hero sha256 matches.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Charged amount is $0.
Footage binding: sha256_pinned
gpt-6.1-sol · OpenAI Codex · Codex CLI 0.160.0 · Max
HERO rendered captureUnaided model$17.39 API-equivalent estimate93m 17sLedger outcome: FAILTask judge: fail
Record details
Existing public result · Ledger status: judge_fail_independently_replayed
Judge fail. Cycles-hero sha256 matches. The hero does not change the fail.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-opus-5-5 · Anthropic Claude Code · Claude CLI 2.1.280 · Max
HERO rendered captureUnaided model$13.09 API-equivalent estimate1h 39m 54sLedger outcome: FAILTask judge: fail
Record details
Canonical harness projection · Ledger status: judge_fail_independently_replayed
Judge fail. Cycles-hero sha256 matches. The hero does not change the fail.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-sonnet-5-5 · Anthropic Claude Code · Claude CLI 2.1.286 · Max
Rubik's cube, single turn
The episode says all four pass, Sol is cheapest, and the Claude runs take longer while staying under 15 minutes. All four unaided max ledgers judge-pass. Sol's API-equivalent is $0.25. Opus is 9m 41s and Sonnet is 11m 25s.
Renderer-bound heroUnaided model$1.86 API-equivalent estimate4m 19sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass. Preferred picture is the operator-rendered hero sha256 a285f316e21c051b0040d835f684bb9a78b525c001b96e5a6aa38593dfbf7682, 170 frames at 30 fps. ctrl.npy 8b5ffd090a7abcda1ff496bad22f0132381873e395f3cbede10a03ed9d4876c5 matches the canonical ledger. This is not a clips.public.json cycles-hero pin. The canonical public clip remains browser sha256 000dcc7984dd25d417524dd15e25b666c4197cbd38f445aa5718ba68179647ae.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Marginal subscription charge is $0. Footage role is operator-renderer binding.
Footage binding: sha256_pinned
gpt-6-astra · OpenAI Codex · Codex CLI 0.154.0 · Max
HERO rendered captureUnaided model$0.25 API-equivalent estimate4m 29sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass. Cycles-hero sha256 matches. This is the lowest API-equivalent of the four unaided max ledgers.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Charged amount is $0.
Footage binding: sha256_pinned
gpt-6.1-sol · OpenAI Codex · Codex CLI 0.160.0 · Max
HERO rendered captureUnaided model$2.20 API-equivalent estimate9m 41sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass. Cycles-hero sha256 matches.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-opus-5-5 · Anthropic Claude Code · Claude CLI 2.1.280 · Max
HERO rendered captureUnaided model$1.50 API-equivalent estimate11m 25sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass. Cycles-hero sha256 matches.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-sonnet-5-5 · Anthropic Claude Code · Claude CLI 2.1.286 · Max
Rubik's cube, three turns
The episode says all four pass, Sol takes 2 hours 31 minutes and is still the cheapest, Sonnet is the most expensive, and Astra costs more than Opus. Sol's display is 151m 49s and the API-equivalent is $3.49. Sonnet is $18.19. Astra is $15.97 and Opus is $4.75. All four judge-pass.
Renderer-bound heroUnaided model$15.97 API-equivalent estimate43m 04sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass. API-equivalent is higher than Opus on this task. Preferred picture is the operator-rendered hero sha256 959db683188a23c55e4ff84ef3e6bbc14b226e908591e8a4624e7558828dc6a7, 1617 frames at 30 fps. ctrl.npy 2151970d9c59ff806d1a4e67bfd4b4b2dfe4607dc2e2d2262a5f6f54d374f4b6 matches the canonical ledger. This is not a clips.public.json cycles-hero pin. The canonical public clip remains browser sha256 19d54267ccab4998e0ccf65ee23979603b4b8e718a009ee2d1903da385981e5c.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Marginal subscription charge is $0. Footage role is operator-renderer binding.
Footage binding: sha256_pinned
gpt-6-astra · OpenAI Codex · Codex CLI 0.154.0 · Max
HERO rendered captureUnaided model$3.49 API-equivalent estimate151m 49sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass. 151m 49s matches the spoken 2 hours 31 minutes. This is GPT-6.1 Sol, not GPT-6 Sol. Cycles-hero sha256 matches.
Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Charged amount is $0.
Footage binding: sha256_pinned
gpt-6.1-sol · OpenAI Codex · Codex CLI 0.160.0 · Max
HERO rendered captureUnaided model$4.75 API-equivalent estimate32m 29sLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Judge pass. Cycles-hero sha256 matches. API-equivalent is below Astra on this task.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-opus-5-5 · Anthropic Claude Code · Claude CLI 2.1.280 · Max
HERO rendered captureUnaided model$18.19 API-equivalent estimate1h 40m 47sLedger outcome: PASSTask judge: pass
Record details
Canonical harness projection · Ledger status: judge_pass_independently_replayed
Judge pass. This is the highest API-equivalent of the four unaided max ledgers. Cycles-hero sha256 matches.
Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
Footage binding: sha256_pinned
claude-sonnet-5-5 · Anthropic Claude Code · Claude CLI 2.1.286 · Max
JRPG boss battle
The episode says the games are functional, Sol is cheapest, Sonnet costs about $53 on max and runs past two hours, and Opus is cheap. Sol's API-equivalent is $1.61 with verdict QUALIFIED and status session_finished_pending_acceptance. Sonnet's API-equivalent is $53.50 and the status remains partial_user_stopped. Opus's deduplicated API-equivalent is $6.17. The spoken $61 is not the Opus ledger figure. Astra's display is 34m 22s and the API-equivalent is $12.65.
Supplied source captureUnaided model$12.65 API-equivalent estimate34m 22sLedger outcome: partial_completed_run_metrics_with_live_operator_smoke
Record details
Existing public result · Ledger status: partial_completed_run_metrics_with_live_operator_smoke
Partial completed-run metrics with a live operator smoke. Not a judge pass.
Wall clock is supplied_measurements.wall_clock_display. Cost is api_cost.total_usd.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · Max
Supplied source captureUnaided model$6.17 API-equivalent estimate34m 31sLedger outcome: complete
Record details
Existing public result · Ledger status: complete
Status complete. Deduplicated API-equivalent is $6.17. The spoken $61 is not this ledger figure.
Clock is measurement.generation_wall_seconds. Cost is primary_claude_api_list_price_equivalent_usd. The uncorrected metrics-prompt total is not used.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · High
Supplied source captureUnaided model$53.50 API-equivalent estimate2h 03m 03sLedger outcome: PARTIAL
Record details
Canonical harness projection · Ledger status: partial_user_stopped
partial_user_stopped and verdict PARTIAL. API-equivalent $53.50 matches the spoken figure. The episode calling the game functional does not clear this status.
Clock is timing.elapsed_seconds_including_quota_and_queue_waits. Cost is recorded_cost.api_equivalent_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Supplied source captureUnaided model$1.61 API-equivalent estimate47m 05sLedger outcome: QUALIFIED
Record details
Canonical harness projection · Ledger status: session_finished_pending_acceptance
Verdict QUALIFIED while acceptance is pending. This is the lowest API-equivalent of the four rows. Not recorded as a judge pass. Not the older GPT-6 Sol ledger.
Clock is measurements.model_invocation_wall_seconds. Cost is api_cost.total_usd. Actual subscription charge is null.
Footage binding: operator_supplied
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Neon Gunner
The episode says Sol costs $1.77 and takes 1 hour 9 minutes, Astra takes 44 minutes, and the Claude runs take about two hours. Sol's API-equivalent is $1.77 and the invocation wall is 1h 9m 50s, with verdict QUALIFIED. Astra's display is 44m 55.2s. Opus's quota-adjusted active workflow is 2h 1m 19s. Sonnet's elapsed time is 2h 1m 42s and the status remains partial_user_stopped.
Supplied source captureUnaided model$9.90 API-equivalent estimate44m 55.2s (supplemental snapshot span)Ledger outcome: partial_post_task_metrics_snapshot_with_supplied_validation
Record details
Existing public result · Ledger status: partial_post_task_metrics_snapshot_with_supplied_validation
Partial post-task snapshot. The display includes metrics follow-up. Not a judge pass.
Wall clock is supplied_measurements.wall_clock_seconds 2695.168. Display is 44m 55.2s (supplemental snapshot span). Cost is api_cost.total_usd 9.899382.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · Max
Supplied source captureUnaided model$33.38 API-equivalent estimate2h 01m 19sLedger outcome: complete
Record details
Existing public result · Ledger status: complete
Status complete. The selected clock excludes a quota pause. Not recorded as a separate judge pass.
Clock is measurement.quota_adjusted_active_workflow_seconds. Chronological time including a 7863.746-second quota pause is 15143.228 seconds and is not the selected clock. Cost is the deduplicated primary-session API-equivalent, not the uncorrected prompt total.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Supplied source captureUnaided model$45.16 API-equivalent estimate2h 01m 42sLedger outcome: PARTIAL
Record details
Canonical harness projection · Ledger status: partial_user_stopped
partial_user_stopped and verdict PARTIAL. The about-two-hour narration does not clear this status.
Clock is timing.elapsed_seconds_including_quota_and_queue_waits. Cost is recorded_cost.api_equivalent_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Supplied source captureUnaided model$1.77 API-equivalent estimate1h 09m 50sLedger outcome: QUALIFIED
Record details
Canonical harness projection · Ledger status: session_finished_pending_acceptance
API-equivalent $1.77 and about 1h 9m match the episode. Verdict QUALIFIED while acceptance is pending, so validationPassed is unset.
Clock is measurements.model_invocation_wall_seconds 4189.563. Cost is api_cost.total_usd. Actual subscription charge is null.
Footage binding: operator_supplied
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Sekiro main menu
The episode says the GPT models use an image-generation shortcut while Claude builds the scene in-engine, and that this is why the cost and time differ. Opus max is partial: 4h 14m 39.541s and $93.87 deduplicated API-equivalent, with the contestant visual review failed. A later high-effort Opus attempt is complete at 46m 30s and $11.17 and stays related, because the large cost gap matches the max run. Sol is $1.57 with verdict QUALIFIED. Sonnet is partial_user_stopped at $70.13.
Supplied source captureUnaided model$11.65 API-equivalent estimate44m 15.4s (qualified snapshot span)Ledger outcome: partial_post_task_metrics_snapshot_with_build_and_live_load
Record details
Existing public result · Ledger status: partial_post_task_metrics_snapshot_with_build_and_live_load
Partial snapshot. The episode's image-generation shortcut remark is editorial and is not a second measured cost.
Wall clock is supplied_measurements.wall_clock_display. workflow seconds are wall_clock_seconds 2655.423. Cost is api_cost.total_usd.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · Max
Supplied source captureUnaided model$93.87 API-equivalent estimate4h 14m 39.541sLedger outcome: partial
Record details
Existing public result · Ledger status: partial
Partial max run. The contestant visual review failed. This is the long, expensive in-engine ledger. A later complete high attempt is retained. The published clip is the operator-supplied Opus Sekiro capture. The filename does not say max or high, and the hash is not a canonical clips pin.
Clock is measurement.generation_wall_display. Cost is primary_session_api_list_price_equivalent_usd for the deduplicated primary session. The uncorrected exact-prompt total is not used.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Retained retry · complete · Later high-effort completion at 2790.154 seconds and API-equivalent $11.17. It is not the long max run.
Supplied source captureUnaided model$70.13 API-equivalent estimate2h 59m 47sLedger outcome: PARTIAL
Record details
Canonical harness projection · Ledger status: partial_user_stopped
partial_user_stopped and verdict PARTIAL.
Clock is timing.elapsed_seconds_including_quota_and_queue_waits. CLI-reported session seconds are 10734.339. Cost is recorded_cost.api_equivalent_usd. Actual subscription charge is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · Max
Supplied source captureUnaided model$1.57 API-equivalent estimate51m 06sLedger outcome: QUALIFIED
Record details
Canonical harness projection · Ledger status: session_finished_pending_acceptance
Verdict QUALIFIED while acceptance is pending. Not the older GPT-6 Sol menu ledger.
Clock is measurements.model_invocation_wall_seconds. Cost is api_cost.total_usd. Actual subscription charge is null.
Footage binding: operator_supplied
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
ATS depot driving
The episode shows Opus on high missing the turn inside the depot, then Astra on high reaching the road and then taking the wrong lane movement before the run is stopped. Neither completes the route. Sonnet has a same-window ledger and is not narrated: it stays in the depot. There is no GPT-6.1 Sol ledger. All three statuses are historical_exploratory_import_incomplete. No route is recorded as passed.
Supplied source captureUnaided modelNot recorded50m 13sLedger outcome: historical_exploratory_import_incomplete
Record details
Canonical harness projection · Ledger status: historical_exploratory_import_incomplete
Reached the public road and then recorded a lane-use error. The user stopped the run. Route not completed. The published clip is the operator-supplied high gameplay capture. A second Astra driving file is not the published derivative.
Clock is supplied_measurements.active_wall_seconds, excluding repair downtime. billed_cost_usd and estimated_cost_usd are null.
Footage binding: operator_supplied
gpt-6-astra · OpenAI Codex · Codex CLI · High
Supplied source captureUnaided model$6.82 API-equivalent estimate10m 21sLedger outcome: historical_exploratory_import_incomplete
Record details
Canonical harness projection · Ledger status: historical_exploratory_import_incomplete
Stayed in the depot after a harness repair. Route not completed. Not narrated as a pass. The estimate is not a billed charge.
Clock is supplied_measurements.active_wall_seconds. Cost is supplied_measurements.cost.estimated_cost_usd. billed_cost_usd is null.
Footage binding: operator_supplied
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · High
Supplied source captureUnaided model$0.45 API-equivalent estimate2m 56sLedger outcome: historical_exploratory_import_incomplete
Record details
Canonical harness projection · Ledger status: historical_exploratory_import_incomplete
Same-window attempt. The episode does not narrate Sonnet. The run stayed in the depot. Route not completed.
Clock is supplied_measurements.active_wall_seconds. Cost is supplied_measurements.cost.estimated_cost_usd. billed_cost_usd is null.
Footage binding: operator_supplied
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · High
CAD-002 robot arm
The supplied robot-arm clips are CAD-002, not the turbofan chapter and not pick-and-place. Opus high is scored_complete. Sonnet high is scored_incomplete. GPT-6.1 Sol max is scored_complete. No unaided Astra CAD-002 ledger was found. scored_complete is not recorded as a motion-contract pass.
Supplied source captureUnaided model$9.49 API-equivalent estimate29m 47sLedger outcome: scored_complete
Record details
Campaign evidence · Ledger status: scored_complete
Supplemental CAD-002 robot arm, fixture cad-002-robot-arm-v1.0.1. scored_complete is not a motion-contract pass. Distinct from CAD-001 and from pick-and-place.
Clock is model_wall_clock_seconds. Cost is cli_list_rate_equivalent_usd. actual_subscription_charge_usd is null. The mp4 sha256 matches the operator comparison artifact.
Footage binding: sha256_pinned
claude-opus-5-5 · Anthropic Claude Code · Claude Code CLI · High
Supplied source captureUnaided model$8.89 API-equivalent estimate23m 04sLedger outcome: scored_incomplete
Record details
Campaign evidence · Ledger status: scored_incomplete
Supplemental. scored_incomplete. The list-rate total includes a small Haiku helper component. Not a motion-contract pass.
Clock is model_wall_clock_seconds. Cost is cli_list_rate_equivalent_usd. actual_subscription_charge_usd is null. The mp4 sha256 matches the operator comparison artifact.
Footage binding: sha256_pinned
claude-sonnet-5-5 · Anthropic Claude Code · Claude Code CLI · High
Supplied source captureUnaided modelAPI-equivalent range $2.75–$5.1953m 48sLedger outcome: scored_complete
Record details
Campaign evidence · Ledger status: scored_complete
Supplemental. scored_complete is not a motion-contract pass. The 2026-10-02 pilot is not this row. Cost is a range.
Clock is model_wall_clock_seconds. Cost is api_equivalent_cost_range_usd [2.752497, 5.192219]. actual_subscription_charge_usd is null. The mp4 sha256 matches the operator comparison artifact.
Footage binding: sha256_pinned
gpt-6.1-sol · OpenAI Codex · Codex CLI · Max
Eiffel drawing, Astra planner plus Jev
The supplied Eiffel clip is this planner-plus-Jev workflow. The unaided four-model chapter is separate. The judge passes this supplemental run. Its cycles-hero file matches the pin. The supplied original in the four-file folder does not match the pinned browser clip.
HERO rendered captureAstra planner + Jev · not the unaided episode attempt$0.18 API-equivalent estimateLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Supplemental Astra low planner plus Jev 1.13. Judge pass. Cycles-hero sha256 matches. The supplied original does not match the pinned browser clip and is not the published picture.
Cost is api_cost.api_equivalent_usd. Wall clock is left unset on this supplemental row.
Footage binding: sha256_pinned
astra-planner-jev · OpenAI Codex
Apple carry, Astra planner plus Jev
The supplied apple-carry clip is this planner-plus-Jev workflow. The pick-and-place chapter uses the unaided max runs. This supplemental capture matches the pinned non-hero clip. Its cycles-hero commitment is not a rendered file.
Recorded browser captureAstra planner + Jev · not the unaided episode attempt$3.83 API-equivalent estimateLedger outcome: PASSTask judge: pass
Record details
Existing public result · Ledger status: judge_pass_independently_replayed
Supplemental Astra low planner plus Jev 1.13 on fr3-gripper. The supplied file matches the pinned non-hero clip. The cycles-hero commitment is not a rendered file.
Cost is api_cost.api_equivalent_usd. Wall clock is left unset on this supplemental row. This is not the unaided pick-and-place column.
Footage binding: sha256_pinned
astra-planner-jev · OpenAI Codex
Method
- The 20 episode rounds are the published chapters. Supplemental footage is not an extra chapter.
- Astra planner + Jev clips are a different workflow from the unaided model attempts.
- Validation status comes from the ledger. Episode narration does not replace it.
- Missing cost, wall clock, and pass/fail stay blank.
- Inline footage plays only when the clip key and result id match one hash-bound public file.
- Robot captures are simulated. This episode records no physical trials.
- Games and visual artifacts were generated. Doom and ATS control existing simulations.
- Differing efforts and interventions are not one controlled single-configuration cohort.
- API-equivalent amounts are list-rate estimates, not billed charges.
- There is no aggregate score and no new blind-vote cohort.
- Claude Opus 5.5
- Unaided episode attempts use the Opus 5.5 lineage recorded on each ledger.
- GPT-6 Astra
- Unaided Astra attempts are not the low-effort or planner-plus-Jev lineage.
- Claude Sonnet 5.5
- Unaided episode attempts use the Sonnet 5.5 lineage recorded on each ledger.
- GPT-6.1 Sol
- GPT-6 Sol, without the 6.1 designation, is a different lineage and is not substituted.
- Astra planner + Jev
- Supplemental workflow. It is not the unaided episode attempt for the same task.
- Hero footage
- A rendered clip is shown only when its published file is hash-bound to that exact result.