DROPGoogle launches Gemini 3.8 Flash TTS and Flash-Lite TTS///DROPGemini Omni brings 1080p generation to Google Vids///BREAKClaude Opus 5.5 is here: 1M context, $4 input / $20 output///BREAKOpenAI adds GPT-6 Sol and Luna to the GPT-6 family///PAPEROpus 5.5's system card documents gains—and evaluation limits///BREAKGrok 4.7 launches at $2 input / $6 output per million tokens///DROPGrok Voice Transcribe 2.0 targets difficult real-world audio///TRENDGoogle steers new Gemini API projects to 3.8 Flash or 3.5 Flash-Lite///DROPGoogle launches Gemini 3.8 Flash TTS and Flash-Lite TTS///DROPGemini Omni brings 1080p generation to Google Vids///BREAKClaude Opus 5.5 is here: 1M context, $4 input / $20 output///BREAKOpenAI adds GPT-6 Sol and Luna to the GPT-6 family///PAPEROpus 5.5's system card documents gains—and evaluation limits///BREAKGrok 4.7 launches at $2 input / $6 output per million tokens///DROPGrok Voice Transcribe 2.0 targets difficult real-world audio///TRENDGoogle steers new Gemini API projects to 3.8 Flash or 3.5 Flash-Lite///
DARC.DEV

[ MODEL — openai / DC Model Index ]

GPT-6 Luna

GPT-6 Luna is the fast, cost-efficient model in OpenAI's GPT-6 series, positioned below GPT-6 Sol. It is suited for high-volume and latency-sensitive workloads such as chat, classification, and lightweight agentic...

Update delayed · last successful import · 437 models · Sources & methodology ↗

openai · Released 2026-09-22

GPT-6 Luna
51.7

DC Power

100% fixed panel · 4 source families
47.9

DC Eco

50.8

DC Overall

Coding

55.6

Full panel

Reasoning

71.9

Full panel

Agents

23.6

Full panel

1.05M context · $0.1 / $0.5

USD per 1M input / output tokens

GPT-6 Luna is the fast, cost-efficient model in OpenAI's GPT-6 series, positioned below GPT-6 Sol. It is suited for high-volume and latency-sensitive workloads such as chat, classification, and lightweight agentic...

Evidence & available offers →

Available offers & pricing

Catalogue listings are not a guarantee of deployability. Azure quota tier, regional capacity and account eligibility require a separate account-specific check. No account information is included here.

ProviderInput / output per 1MRegion / deploymentQuota tier
OpenRouter ↗openai/gpt-6-luna$0.1 / $0.5Not verified / On-demandAccount-specific
Azure ↗gpt-6-luna$0.1 / $0.5Not verified / Not verifiedAccount-specific
openai ↗gpt-6-luna$0.1 / $0.5Not verified / Not verifiedAccount-specific

Lowest listed Azure retail regions

For the reference workload of 3,000 input and 1,000 output tokens. Each group compares only matching deployment and processing options. Global deployments can process data outside the resource region. Data Zone regions may belong to different residency zones; check your requirements before choosing. Catalogue pricing does not confirm quota, model access or regional capacity.

  • Global Standard · Flex · short context: $0.0004 per task · australiaeast, brazilsouth, canadaeast, centralus, eastus, eastus2, francecentral, germanywestcentral, indonesiacentral, japaneast, koreacentral, malaysiawest, northcentralus, norwayeast, polandcentral, southafricanorth, southcentralus, southindia, spaincentral, swedencentral, switzerlandnorth, uaenorth, uksouth, westeurope, westus, westus3
  • Global Standard · Standard · short context: $0.0008 per task · australiaeast, brazilsouth, canadaeast, centralus, eastus, eastus2, francecentral, germanywestcentral, indonesiacentral, japaneast, koreacentral, malaysiawest, northcentralus, norwayeast, polandcentral, southafricanorth, southcentralus, southindia, spaincentral, swedencentral, switzerlandnorth, uaenorth, uksouth, westeurope, westus, westus3
  • Data Zone Standard · Standard · short context: $0.00088 per task · centralus, eastus, eastus2, northcentralus, southcentralus, westus, westus3

Reference task estimate: $0.0008 for 3,000 input + 1,000 output tokens. Cache reads: $0.01 / 1M tokens where listed.

Supplementary ECI index · 0% of Power

Epoch Capabilities Index: 156.68 · published 90% interval 154.78–158.74

18 ECI benchmarks; mixed source configurations. Overlapping intervals do not establish superiority or equivalence.

Epoch AI methodology & data ↗
Use GPT-6 Luna as the comparison reference →
Inspect GPT-6 Luna: weights, inputs & missing evidence

Fixed-panel coverage: 100%. All ranked models use this complete panel. ECI contributes 0%.

Coding: 55.6 · 35% of Power · 100% coverage

LiveBench 2 Coding · 25 Oct 2026 · 50% of Coding = 17.5% of Power · 84.5

LiveBench 2 Coding · 20261025: 84.50% ↗gpt-6-luna · max · LiveBench 2 20261025 · LiveBench 2 · 20261025 · LiveBench

Epoch AI original MirrorCode · 50% of Coding = 17.5% of Power · 26.7

MirrorCode: 26.67% ↗gpt-6-luna_max · Epoch Inspect · Epoch AI evaluation

Reasoning: 71.9 · 35% of Power · 100% coverage

LiveBench 2 Reasoning · 25 Oct 2026 · 40% of Reasoning = 14.0% of Power · 81.8

LiveBench 2 Reasoning · 20261025: 81.80% ↗gpt-6-luna · max · LiveBench 2 20261025 · LiveBench 2 · 20261025 · LiveBench

Epoch AI original reasoning panel · 30% of Reasoning = 10.5% of Power · 71.3

Chess Puzzles: 36.00% ↗gpt-6-luna_max · Epoch Inspect · Epoch AI evaluation · each input 33.33% of this panel

FrontierMath-Tiers-1-3-v2-Private: 78.95% ↗gpt-6-luna_max · Epoch Inspect · Epoch AI evaluation · each input 33.33% of this panel

OTIS Mock AIME 2024-2025: 98.89% ↗gpt-6-luna_max · Epoch Inspect · Epoch AI evaluation · each input 33.33% of this panel

ARC Prize · ARC-AGI-2 semi-private · 30% of Reasoning = 10.5% of Power · 59.3

ARC-AGI-2 · semi-private: 59.31% ↗GPT-6 Luna (Max) · ARC Prize leaderboard baseline · ARC Prize official leaderboard

Agents: 23.6 · 30% of Power · 100% coverage

LiveBench 2 Long Agentic · 25 Oct 2026 · 50% of Agents = 15.0% of Power · 30.9

LiveBench 2 Long Agentic · 20261025: 30.90% ↗gpt-6-luna · max · LiveBench 2 20261025 · LiveBench 2 · 20261025 · LiveBench

Terminal-Bench 4.0 · published model + agent · 50% of Agents = 15.0% of Power · 16.4

Terminal-Bench 4.0: 16.36% ↗GPT-6 Luna · Codex · max · Terminal-Bench 4.0 · Codex · Terminal-Bench public leaderboard

Power = 0.35 × 55.6 + 0.35 × 71.9 + 0.30 × 23.6 = 51.7. Full precision is used before final rounding.

Original benchmark observations

27 benchmarks · 47 observations. Each configuration stays separate. These are original percentages, before ECI chance and ceiling adjustments. Expand a score to inspect its source and conditions.

BenchmarkScoreEffort / harnessSource
GPQA diamondKnowledge
90.47%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-09-22 · gpqa_diamond.csv

Reported standard error: ±1.85 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
FrontierMath-Tiers-1-3-v2-PrivateReasoning
78.95%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-09-22 · frontiermath_tiers_1_3_v2.csv

Reported standard error: ±2.42 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
FrontierMath-Tier-4-v2-PrivateReasoning
56.10%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-09-22 · frontiermath_tier_4_v2.csv

Reported standard error: ±7.85 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
OTIS Mock AIME 2024-2025Reasoning
98.89%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-09-22 · otis_mock_aime_2024_2025.csv

Reported standard error: ±0.67 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
SimpleQA VerifiedKnowledge
41.40%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-09-22 · simpleqa_verified.csv

Reported standard error: ±1.56 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
Chess PuzzlesReasoning
36.00%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-10-13 · chess_puzzles.csv

Reported standard error: ±4.82 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
Mystery Game PuzzlesReasoning
7.00%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-09-22 · mystery_game_puzzles.csv

Reported standard error: ±2.56 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
MirrorCodeCoding
26.67%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-10-21 · mirrorcode.csv

Reported standard error: ±9.03 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
ARC-AGIReasoning
86.67%

gpt-6-luna_max

https://arcprize.org/leaderboard · external

Run: Date not supplied · arc_agi_external.csv

Inspect source ↗

CC BY 4.0

max / Not reportedExternal · via Epoch ↗
ARC-AGIReasoning
73.00%

gpt-6-luna_xhigh

https://arcprize.org/leaderboard · external

Run: Date not supplied · arc_agi_external.csv

Inspect source ↗

CC BY 4.0

xhigh / Not reportedExternal · via Epoch ↗
ARC-AGIReasoning
70.33%

gpt-6-luna_high

https://arcprize.org/leaderboard · external

Run: Date not supplied · arc_agi_external.csv

Inspect source ↗

CC BY 4.0

high / Not reportedExternal · via Epoch ↗
ARC-AGIReasoning
61.00%

gpt-6-luna_medium

https://arcprize.org/leaderboard · external

Run: Date not supplied · arc_agi_external.csv

Inspect source ↗

CC BY 4.0

medium / Not reportedExternal · via Epoch ↗
ARC-AGIReasoning
37.67%

gpt-6-luna_low

https://arcprize.org/leaderboard · external

Run: Date not supplied · arc_agi_external.csv

Inspect source ↗

CC BY 4.0

low / Not reportedExternal · via Epoch ↗
ARC-AGIReasoning
8.83%

gpt-6-luna_none

https://arcprize.org/leaderboard · external

Run: Date not supplied · arc_agi_external.csv

Inspect source ↗

CC BY 4.0

none / Not reportedExternal · via Epoch ↗
APEX-AgentsAgents
44.30%

gpt-6-luna_max

External result; evaluator not specified · external

Run: Date not supplied · apex_agents_external.csv

Inspect source ↗

CC BY 4.0

max / Not reportedExternal · via Epoch ↗
ARC-AGI-2Reasoning
59.31%

gpt-6-luna_max

External result; evaluator not specified · external

Run: Date not supplied · arc_agi_2_external.csv

Inspect source ↗

CC BY 4.0

max / Not reportedExternal · via Epoch ↗
ARC-AGI-2Reasoning
41.94%

gpt-6-luna_xhigh

External result; evaluator not specified · external

Run: Date not supplied · arc_agi_2_external.csv

Inspect source ↗

CC BY 4.0

xhigh / Not reportedExternal · via Epoch ↗
ARC-AGI-2Reasoning
31.39%

gpt-6-luna_high

External result; evaluator not specified · external

Run: Date not supplied · arc_agi_2_external.csv

Inspect source ↗

CC BY 4.0

high / Not reportedExternal · via Epoch ↗
ARC-AGI-2Reasoning
18.06%

gpt-6-luna_medium

External result; evaluator not specified · external

Run: Date not supplied · arc_agi_2_external.csv

Inspect source ↗

CC BY 4.0

medium / Not reportedExternal · via Epoch ↗
ARC-AGI-2Reasoning
4.58%

gpt-6-luna_low

External result; evaluator not specified · external

Run: Date not supplied · arc_agi_2_external.csv

Inspect source ↗

CC BY 4.0

low / Not reportedExternal · via Epoch ↗
ARC-AGI-2Reasoning
0.00%

gpt-6-luna_none

External result; evaluator not specified · external

Run: Date not supplied · arc_agi_2_external.csv

Inspect source ↗

CC BY 4.0

none / Not reportedExternal · via Epoch ↗
CritPtOther
19.43%

gpt-6-luna_max

External result; evaluator not specified · external

Run: Date not supplied · critpt_external.csv

Inspect source ↗

CC BY 4.0

max / Not reportedExternal · via Epoch ↗
CritPtOther
17.43%

gpt-6-luna_xhigh

External result; evaluator not specified · external

Run: Date not supplied · critpt_external.csv

Inspect source ↗

CC BY 4.0

xhigh / Not reportedExternal · via Epoch ↗
CritPtOther
15.43%

gpt-6-luna_high

External result; evaluator not specified · external

Run: Date not supplied · critpt_external.csv

Inspect source ↗

CC BY 4.0

high / Not reportedExternal · via Epoch ↗
CritPtOther
10.57%

gpt-6-luna_medium

External result; evaluator not specified · external

Run: Date not supplied · critpt_external.csv

Inspect source ↗

CC BY 4.0

medium / Not reportedExternal · via Epoch ↗
CritPtOther
2.57%

gpt-6-luna_low

External result; evaluator not specified · external

Run: Date not supplied · critpt_external.csv

Inspect source ↗

CC BY 4.0

low / Not reportedExternal · via Epoch ↗
CritPtOther
1.14%

gpt-6-luna_none

External result; evaluator not specified · external

Run: Date not supplied · critpt_external.csv

Inspect source ↗

CC BY 4.0

none / Not reportedExternal · via Epoch ↗
ProofBenchReasoning
64.00%

gpt-6-luna_unknown

https://www.vals.ai/benchmarks/proof_bench · external

Run: Date not supplied · proofbench_external.csv

Inspect source ↗

CC BY 4.0

unknown / Not reportedExternal · via Epoch ↗
FrontierCodeCoding
42.42%

gpt-6-luna_max

https://cognition.com/frontiercode · external

Run: Date not supplied · frontiercode_external.csv

Inspect source ↗

CC BY 4.0

max / codexExternal · via Epoch ↗
LMCAReasoning
44.49%

gpt-6-luna_max

https://conceptualreasoning.ai/lmca · external

Run: Date not supplied · lmca_external.csv

Inspect source ↗

CC BY 4.0

max / Not reportedExternal · via Epoch ↗
DTBenchReasoning
90.13%

gpt-6-luna_max

https://conceptualreasoning.ai/dtbench · external

Run: Date not supplied · dtbench_external.csv

Inspect source ↗

CC BY 4.0

max / Not reportedExternal · via Epoch ↗
Furniture AssemblyReasoning
44.17%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-09-28 · furniture_assembly.csv

Reported standard error: ±6.41 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
GDP.pdfOther
23.00%

gpt-6-luna_max

https://surgehq.ai/benchmarks/gdp-pdf · external

Run: Date not supplied · gdp_pdf_external.csv

Inspect source ↗

CC BY 4.0

max / Not reportedExternal · via Epoch ↗
WeirdML v3Coding
12.02%

gpt-6-luna_xhigh

https://htihle.github.io/weirdml.html · external

Run: Date not supplied · weirdml_v3_external.csv

Inspect source ↗

CC BY 4.0

xhigh / codex_cliExternal · via Epoch ↗
SpireClimb (Slay the Spire 1, Ascension 20 Heart)Other
0.00%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-11-05 · spireclimb_sts1_a20_heart.csv

Reported standard error: ±0.00 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
SpireClimb (Slay the Spire 2, Ascension 10)Other
0.00%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-09-28 · spireclimb_sts2_a10.csv

Reported standard error: ±0.00 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
SpireClimbOther
0.00%

gpt-6-luna_max

Epoch AI evaluation · epoch-run

Run: 2026-11-05 · spireclimb.csv

Reported standard error: ±0.00 pp. This is not a 90% interval.

Inspect source ↗

CC BY 4.0

max / Epoch InspectEpoch AI run ↗
LiveBench 2 Coding · 20261025Coding
84.50%

gpt-6-luna · max · LiveBench 2 20261025

LiveBench · independent-run

Run: Date not supplied · 20261025/table.csv

Benchmark edition 25 October 2026; individual run date not supplied. One run per model; mean task score. Failed/unmeasurable tasks score zero in the source. Source effort metadata is preserved.

Inspect source ↗

Attributed public leaderboard score facts · LiveBench

max / LiveBench 2 · 20261025LiveBench ↗
LiveBench 2 Reasoning · 20261025Reasoning
81.80%

gpt-6-luna · max · LiveBench 2 20261025

LiveBench · independent-run

Run: Date not supplied · 20261025/table.csv

Benchmark edition 25 October 2026; individual run date not supplied. One run per model; mean task score. Failed/unmeasurable tasks score zero in the source. Source effort metadata is preserved.

Inspect source ↗

Attributed public leaderboard score facts · LiveBench

max / LiveBench 2 · 20261025LiveBench ↗
LiveBench 2 Long Agentic · 20261025Agents
30.90%

gpt-6-luna · max · LiveBench 2 20261025

LiveBench · independent-run

Run: Date not supplied · 20261025/table.csv

Benchmark edition 25 October 2026; individual run date not supplied. One run per model; mean task score. Failed/unmeasurable tasks score zero in the source. Source effort metadata is preserved.

Inspect source ↗

Attributed public leaderboard score facts · LiveBench

max / LiveBench 2 · 20261025LiveBench ↗
Terminal-Bench 4.0Agents
16.36%

GPT-6 Luna · Codex · max

Terminal-Bench public leaderboard · benchmark-publisher

Run: 2026-09-22 · 4-0-0/5bd8ac99-7353-4210-bdc5-153029e660d5

Published model + agent result; agents differ across models. Not a controlled model-only evaluation. Reported trials: 330. Source 95% accuracy interval half-width: 2.72 pp; this is not a DC Power interval.

Inspect source ↗

Attributed public leaderboard score facts · Terminal-Bench

max / Terminal-Bench 4.0 · CodexTerminal-Bench ↗
ARC-AGI-2 · semi-privateReasoning
59.31%

GPT-6 Luna (Max)

ARC Prize official leaderboard · benchmark-publisher

Run: Date not supplied · evaluations.json · v2_Semi_Private

Reported benchmark cost: $0.061811 per task. Different from the token-workload estimate.

Semi-private split only. Provider-adapter and custom-scaffold families are kept separate. Identical effort names do not imply identical compute budgets.

Inspect source ↗

Public leaderboard facts · ARC Prize

max / ARC Prize leaderboard baselineARC Prize ↗
ARC-AGI-2 · semi-privateReasoning
41.94%

GPT-6 Luna (XHigh)

ARC Prize official leaderboard · benchmark-publisher

Run: Date not supplied · evaluations.json · v2_Semi_Private

Reported benchmark cost: $0.032298 per task. Different from the token-workload estimate.

Semi-private split only. Provider-adapter and custom-scaffold families are kept separate. Identical effort names do not imply identical compute budgets.

Inspect source ↗

Public leaderboard facts · ARC Prize

xhigh / ARC Prize leaderboard baselineARC Prize ↗
ARC-AGI-2 · semi-privateReasoning
31.39%

GPT-6 Luna (High)

ARC Prize official leaderboard · benchmark-publisher

Run: Date not supplied · evaluations.json · v2_Semi_Private

Reported benchmark cost: $0.023362 per task. Different from the token-workload estimate.

Semi-private split only. Provider-adapter and custom-scaffold families are kept separate. Identical effort names do not imply identical compute budgets.

Inspect source ↗

Public leaderboard facts · ARC Prize

high / ARC Prize leaderboard baselineARC Prize ↗
ARC-AGI-2 · semi-privateReasoning
18.06%

GPT-6 Luna (Medium)

ARC Prize official leaderboard · benchmark-publisher

Run: Date not supplied · evaluations.json · v2_Semi_Private

Reported benchmark cost: $0.012203 per task. Different from the token-workload estimate.

Semi-private split only. Provider-adapter and custom-scaffold families are kept separate. Identical effort names do not imply identical compute budgets.

Inspect source ↗

Public leaderboard facts · ARC Prize

medium / ARC Prize leaderboard baselineARC Prize ↗
ARC-AGI-2 · semi-privateReasoning
4.58%

GPT-6 Luna (Low)

ARC Prize official leaderboard · benchmark-publisher

Run: Date not supplied · evaluations.json · v2_Semi_Private

Reported benchmark cost: $0.005274 per task. Different from the token-workload estimate.

Semi-private split only. Provider-adapter and custom-scaffold families are kept separate. Identical effort names do not imply identical compute budgets.

Inspect source ↗

Public leaderboard facts · ARC Prize

low / ARC Prize leaderboard baselineARC Prize ↗
ARC-AGI-2 · semi-privateReasoning
0.00%

GPT-6 Luna (None)

ARC Prize official leaderboard · benchmark-publisher

Run: Date not supplied · evaluations.json · v2_Semi_Private

Reported benchmark cost: $0.003699 per task. Different from the token-workload estimate.

Semi-private split only. Provider-adapter and custom-scaffold families are kept separate. Identical effort names do not imply identical compute budgets.

Inspect source ↗

Public leaderboard facts · ARC Prize

none / ARC Prize leaderboard baselineARC Prize ↗

DC Score v3 / independent source families

One fixed panel. Every weight visible.

DC Power is our weighted aggregate of raw benchmark scores from LiveBench, Epoch AI, ARC Prize and Terminal-Bench. It is not an ECI rescaling. Power = 0.35 × Coding + 0.35 × Reasoning + 0.30 × Agents. These are explicit editorial preferences, not statistically learned weights or a probability of success.

DomainPower weightSource panels within the domain
Coding35%

50% LiveBench 2 Coding · 25 Oct 2026

50% Epoch AI original MirrorCode

Reasoning35%

40% LiveBench 2 Reasoning · 25 Oct 2026

30% Epoch AI original reasoning panel

30% ARC Prize · ARC-AGI-2 semi-private

Agents30%

50% LiveBench 2 Long Agentic · 25 Oct 2026

50% Terminal-Bench 4.0 · published model + agent

Source balance. At full coverage, LiveBench contributes 46.5%, original Epoch AI runs 28%, ARC Prize 10.5% and Terminal-Bench 15%. One source does not gain weight by publishing more rows. ECI, vendor reports and external scores republished through Epoch have zero Power weight. Direct ARC and Terminal results enter once.

Epoch reasoning panel. Equal weights for original Epoch-run Chess Puzzles, FrontierMath tiers 1–3 v2 private and OTIS Mock AIME 2024–2025. All three are required. GPQA and other supplementary tests stay in the matrix; vendor-only GPQA reports do not become independent Epoch runs.

Missing evidence stays visible. Power, Eco and Overall require every fixed source panel. Domain scores can show a partial summary: available panel weights are renormalized, with coverage and missing inputs shown. A partial score is not a win against a complete score. The same complete panel is required for every ranked model.

Fixed editions and effort policy. LiveBench 2 edition 25 October 2026 and Terminal-Bench 4.0. Select known effort in this order: max, xhigh, high, medium, low, minimal, none, declared highest-supported. Never select the largest numerical result. A dated later run can replace an earlier run under the same setup; unresolved conflicts or different harnesses are missing.

Publisher independence is limited. Four separate source organizations do not imply blind or independently reproduced runs, independent funding, equal compute budgets or uncorrelated tests. Terminal-Bench measures a model with its published agent; Claude Code, Codex and other harnesses differ. Inspect the selected configurations. Public benchmark data cannot establish outcomes in your application.

No invented confidence interval. We do not convert ECI uncertainty into uncertainty for this aggregate. No statistically valid DC Power interval is available. Raw means are averaged on their original 0–100 scales; test difficulty and contamination can affect results. A small aggregate difference is descriptive, not proof of superiority.

Token cost C = (3,000 × input rate + 1,000 × output rate) / 1,000,000, uncached. DC Eco = Power / (1 + 100 × C), eligible only for complete Power ≥ 50 and known prices. At $0.01, Eco is half Power. DC Overall = 0.75 × Power + 0.25 × Eco. Use the comparison calculator for your workload and quality weight.

Cost is conditional. Prices are listed USD per million tokens. Reasoning tokens, tools, retries, cache writes, discounts and long-context surcharges are excluded. These are token-spend estimates, not measured cost per successful task. Quality benchmark effort does not establish a matching token workload.

Inspect Claude Fable 5.1: weights, inputs & missing evidence

Fixed-panel coverage: 100%. All ranked models use this complete panel. ECI contributes 0%.

Coding: 85.5 · 35% of Power · 100% coverage

LiveBench 2 Coding · 25 Oct 2026 · 50% of Coding = 17.5% of Power · 97.7

LiveBench 2 Coding · 20261025: 97.70% ↗claude-fable-5.1 · max · LiveBench 2 20261025 · LiveBench 2 · 20261025 · LiveBench

Epoch AI original MirrorCode · 50% of Coding = 17.5% of Power · 73.3

MirrorCode: 73.30% ↗claude-fable-5-1_high · Epoch Inspect · Epoch AI evaluation

Reasoning: 87.5 · 35% of Power · 100% coverage

LiveBench 2 Reasoning · 25 Oct 2026 · 40% of Reasoning = 14.0% of Power · 91.7

LiveBench 2 Reasoning · 20261025: 91.70% ↗claude-fable-5.1 · max · LiveBench 2 20261025 · LiveBench 2 · 20261025 · LiveBench

Epoch AI original reasoning panel · 30% of Reasoning = 10.5% of Power · 79.4

Chess Puzzles: 48.00% ↗claude-fable-5-1_max · Epoch Inspect · Epoch AI evaluation · each input 33.33% of this panel

FrontierMath-Tiers-1-3-v2-Private: 90.18% ↗claude-fable-5-1_max · Epoch Inspect · Epoch AI evaluation · each input 33.33% of this panel

OTIS Mock AIME 2024-2025: 100.00% ↗claude-fable-5-1_max · Epoch Inspect · Epoch AI evaluation · each input 33.33% of this panel

ARC Prize · ARC-AGI-2 semi-private · 30% of Reasoning = 10.5% of Power · 90.0

ARC-AGI-2 · semi-private: 90.00% ↗Claude Fable 5.1 (Max) · ARC Prize leaderboard baseline · ARC Prize official leaderboard

Agents: 67.9 · 30% of Power · 100% coverage

LiveBench 2 Long Agentic · 25 Oct 2026 · 50% of Agents = 15.0% of Power · 77.9

LiveBench 2 Long Agentic · 20261025: 77.90% ↗claude-fable-5.1 · max · LiveBench 2 20261025 · LiveBench 2 · 20261025 · LiveBench

Terminal-Bench 4.0 · published model + agent · 50% of Agents = 15.0% of Power · 57.9

Terminal-Bench 4.0: 57.88% ↗Claude Fable 5.1 · Claude Code · max · Terminal-Bench 4.0 · Claude Code · Terminal-Bench public leaderboard

Power = 0.35 × 85.5 + 0.35 × 87.5 + 0.30 × 67.9 = 80.9. Full precision is used before final rounding.

Reference token cost: $0.08 · Eco: 9.0 · Overall: 62.9

v3 replaces the previous ECI transform; scores across methodology versions are not directly comparable. Data refreshes every six hours, not on each page view. Benchmark editions are pinned for comparability; changing the test panel requires a methodology revision. Imports older than 24 hours are marked delayed. Catalogue releases continue to update independently. Azure listings do not confirm account quota or regional capacity.

OpenRouter catalogue ↗ · Public catalogue API; attributed metadata and list prices

models.dev ↗ · MIT

Epoch AI, Capabilities & benchmarking ↗ · CC BY 4.0; only original Epoch-run panel observations enter Power; ECI has zero weight

ARC Prize official leaderboard ↗ · Attributed public leaderboard facts; semi-private split; one direct Power panel, never counted through Epoch again

LiveBench 2 · 25 October 2026 ↗ · Attributed public score facts; original Coding, Reasoning and Long Agentic suites

Terminal-Bench 4.0 ↗ · Attributed public score facts; published model plus agent, not a model-only ablation

Microsoft Azure Retail Prices ↗ · Public retail meters; deployment eligibility is not confirmed

Epoch data: CC BY 4.0. Other publishers: attributed public leaderboard score facts. DARC.DEV defines the aggregate and makes no claim of source endorsement. No private account details are published.

← Back to DC Model Index