Claude Fable 5 vs GPT-5.6 Luna
Full profiles: Claude Fable 5 (Anthropic) and GPT-5.6 Luna (OpenAI): every benchmark, price and rank.
Updated · Data from 9 sources
As of 26 Oct 2026, Claude Fable 5 (max) leads on quality (69.3 vs 59.9 on the BenchLeader Index), but GPT-5.6 Luna (max) costs 44× less, so it is the better buy unless you need the extra quality. Claude Fable 5 leads in every category we measure both in.
Claude Fable 5 benchmarks max Anthropic
GPT-5.6 Luna benchmarks max OpenAIOpen in the comparison toolWhich to pick for…
The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.
| Use case | Pick | Why | Confidence |
|---|---|---|---|
| Coding | Claude Fable 5 | 74 vs 56 category score; SciCode 61.0% vs 53.6% | High |
| Agents & tool use | Claude Fable 5 | 65 vs 49 category score; Terminal-Bench 44.5% vs 17.3% | High |
| Maths | Claude Fable 5 | 72 vs 65 category score; FrontierMath Tiers 1–3 87.0% vs 82.1% | Medium |
| Reasoning | Claude Fable 5 | 74 vs 63 category score; ARC-AGI-2 89.2% vs 59.6% | High |
| Knowledge | Claude Fable 5 | 81 vs 57 category score; AA-Omniscience 43.3 vs -10.3 | High |
| Long documents | Either | 66 vs 67 category score; AA-LCR 82.3% vs 83.7% | Either |
| Lowest cost | GPT-5.6 Luna | $0.450 vs $20.00 per 1M blended; agentic coding session $0.017 vs $0.800 | – |
| Fastest first answer | Either | 101.8 s (Claude Fable 5 (max)) vs 120.3 s (GPT-5.6 Luna (max)) to the first answer token | – |
| High volume on a budget | Neither flagship | A smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs | – |
Key differences
- Claude Fable 5 (max) leads on quality: 69.3 vs 59.9.
- Claude Fable 5 (max) is stronger in agents & tools, coding, composite, instruction following, knowledge, maths, reasoning.
- GPT-5.6 Luna (max) is stronger in long context, multimodal.
- GPT-5.6 Luna (max) is 44× cheaper ($0.450 vs $20.00 per 1M blended).
- GPT-5.6 Luna (max) streams 1.9× faster (126 vs 67 tokens per second).
At a glance
| Spec | Claude Fable 5 | GPT-5.6 Luna |
|---|---|---|
| Developer | Anthropic | OpenAI |
| Released | 9 Jun 2026 | 9 Jul 2026 |
| Context window (tokens) | 1,000,000 | 1,050,000 |
| Compared at | Claude Fable 5 (max) | GPT-5.6 Luna (max) |
| Open weights | No | No |
| Price in / out / cached input per 1M | $10.00 / $50.00 / $1.00 | $0.200 / $1.20 / $0.020 |
Category by category
Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.
| Category | Claude Fable 5 (max) | GPT-5.6 Luna (max) | Leader |
|---|---|---|---|
| Reasoning | 73.8 | 62.9 | Claude Fable 5 |
| Coding | 73.8 | 56.0 | Claude Fable 5 |
| Agents & tools | 65.0 | 49.0 | Claude Fable 5 |
| Maths | 72.4 | 65.0 | Claude Fable 5 |
| Knowledge | 81.0 | 56.5 | Claude Fable 5 |
| Instruction following | 63.1 | – | not comparable |
| Multimodal | – | 61.6 | not comparable |
| Long context | 66.3 | 67.0 | level |
| Composite | 85.0 | 71.2 | Claude Fable 5 |
Where sources disagree
Across the 37 benchmarks both have, Claude Fable 5 leads 34, GPT-5.6 Luna leads 2 and 1 is a tie. Every overall index that rates both puts Claude Fable 5 ahead.
| Overall index | Claude Fable 5 (max) | GPT-5.6 Luna (max) | Leader |
|---|---|---|---|
| BenchLeader Index | 69.3 | 59.9 | Claude Fable 5 |
| AA Intelligence Index v4.3.2 | 49.6 | 37.3 | Claude Fable 5 |
| Arena Text | 1503† | 1453 | Claude Fable 5 |
Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).
Price, speed and latency
List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page; suite cost is what Artificial Analysis paid to run its whole Intelligence Index.
| Measure | Claude Fable 5 (max) | GPT-5.6 Luna (max) |
|---|---|---|
| Blended price per 1M tokens | $20.00 | $0.450 |
| Input / output per 1M | $10.00 / $50.00 | $0.200 / $1.20 |
| Cached input per 1M | $1.00 | $0.020 |
| Price per task: chat reply | $0.019 | $0.0004 |
| Price per task: summarize a 30-page report | $0.150 | $0.0031 |
| Price per task: code edit | $0.135 | $0.0030 |
| Price per task: agentic coding session | $0.800 | $0.017 |
| Price per task: structured extraction | $0.030 | $0.0006 |
| Cost to run a full benchmark suite | $8.75 | $0.178 |
| Output speed | 67 tok/s | 126 tok/s |
| Time to first answer | 101.8 s | 120.3 s |
| Time to a complete answer | 109.3 s | 124.3 s |
Per token is not per task: on Artificial Analysis' full suite Claude Fable 5 (max) cost $8.75 and GPT-5.6 Luna (max) $0.178, because Claude Fable 5 (max) used about 1.1× as many tokens for the same work.
Each model is compared at its best-ranked reasoning-effort setting. Claude Fable 5 (max) scores 69.3 and takes 102 s to its first answer token; Claude Fable 5 scores 66.6 in 4 s, Claude Fable 5 (high) scores 64.4 in 4 s. GPT-5.6 Luna (max) scores 59.9 and takes 120 s to its first answer token; GPT-5.6 Luna (xhigh) scores 59.7 in 42 s, GPT-5.6 Luna (high) scores 57.0 in 10 s. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.
Every benchmark
130 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.
| Metric | Publisher | Claude Fable 5 (max) | GPT-5.6 Luna (max) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 69.3 | 59.9 |
| Agents & tools score | BenchLeader | 65.0 | 49.0 |
| Coding score | BenchLeader | 73.8 | 56.0 |
| Composite score | BenchLeader | 85.0 | 71.2 |
| Instruction following score | BenchLeader | 63.1 | – |
| Knowledge score | BenchLeader | 81.0 | 56.5 |
| Long context score | BenchLeader | 66.3 | 67.0 |
| Maths score | BenchLeader | 72.4 | 65.0 |
| Reasoning score | BenchLeader | 73.8 | 62.9 |
| Multimodal score | BenchLeader | – | 61.6 |
| Blended price $/M | BenchLeader | $20.00 | $0.450 |
| Output speed | BenchLeader | 67 tok/s | 126 tok/s |
| Time to first answer | BenchLeader | 101.8 s | 120.3 s |
| Context window | BenchLeader | 1,000,000 tokens | 1,050,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 85.9% | 91.6% |
| SimpleBench | SimpleBench | 81.9% | 46.8%† |
| Arena Hard Prompts | Arena (LMArena) | 1529† | 1475† |
| LiveBench 1.0 Reasoning | LiveBench | 89.7% | 85.6% |
| GPQA Diamond (AA) | Artificial Analysis | 92.6% | 91.1% |
| Humanity's Last Exam (AA) | Artificial Analysis | 55.5% | 39.5% |
| GPQA Diamond (Vals) | Vals AI | 93.2%† | 91.7% |
| EnigmaEval | Scale AI SEAL | 39.3%† | – |
| Kagi LLM Benchmark | Kagi LLM Benchmark | 88.8%† | 49.1%† |
| ARC-AGI-1 | ARC Prize | 98.5% | 90.7% |
| ARC-AGI-2 | ARC Prize | 89.2% | 59.6% |
| ARC-AGI-3 | ARC Prize | – | 0.2% |
| CritPt | Artificial Analysis | 28.6% | 20.6% |
| MysteryMechanism | Vals AI | – | 14.4% |
| Chess Puzzles | Epoch AI Benchmarking Hub | 47.0% | 43.0% |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 52.0% | 21.0% |
| LMCA | Epoch AI Benchmarking Hub | 60.3% | 48.5% |
| DTBench | Epoch AI Benchmarking Hub | 98.4% | 88.8% |
| ForecastBench | ForecastBench | 66.0% | – |
| BTF-3 | Epoch AI Benchmarking Hub | 13.0%† | – |
| Coding | |||
| SciCode | SciCode | 61.0% | 53.6% |
| WeirdML | WeirdML | 91.9% | 60.9%† |
| FrontierCode | Cognition | 53.5%† | 39.8%† |
| GSO-Bench | GSO-Bench | 78.4%† | – |
| Arena Coding | Arena (LMArena) | 1551† | 1502† |
Show all 130 benchmarks
| Metric | Publisher | Claude Fable 5 (max) | GPT-5.6 Luna (max) |
|---|---|---|---|
| Coding | |||
| Arena WebDev | Arena (LMArena) | 1624† | 1519† |
| LiveBench 1.0 Coding | LiveBench | 86.0% | 82.9% |
| LiveCodeBench | Vals AI | 89.8%† | – |
| IOI | Vals AI | – | 61.8% |
| SWE-bench (Vals) | Vals AI | 95.0%† | 93.0% |
| Code Migration | Vals AI | 55.1%† | 44.5% |
| ProgramBench | Vals AI | 2.0%† | 0.0% |
| Vibe Code Bench 1-100 | Vals AI | – | 22.6% |
| Vibe Code Bench v1.1 | Vals AI | 90.3%† | 77.1% |
| SWE Atlas: Codebase QnA | Scale AI SEAL | 39.0%† | – |
| SWE Atlas: Refactoring | Scale AI SEAL | 54.8%† | – |
| SWE Atlas: Test Writing | Scale AI SEAL | 55.6%† | – |
| MirrorCode | Epoch AI Benchmarking Hub | 63.9%† | – |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | 95.0%† | 61.9%† |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | 69.7% | 67.2% |
| CursorBench | Cursor | – | 35.9% |
| ALE-Bench | Epoch AI Benchmarking Hub | 2041.3† | 1667.4 |
| FrontierSWE | Epoch AI Benchmarking Hub | 47.0% | – |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 42.4% | 11.6% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 84.6% | 80.9% |
| Startrise Frontend | Startrise | 73.8%† | 65.0%† |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | 44.5% | 17.3% |
| Remote Labor Index | Scale AI / CAIS | 15.8%† | – |
| APEX-Agents | Mercor | 63.6%† | 43.0%† |
| Arena Agent | Arena (LMArena) | 7.8† | – |
| LiveBench Agentic Coding | LiveBench | 62.2% | 48.4% |
| Terminal-Bench Hard | Artificial Analysis | 62.9% | – |
| τ²-Bench Telecom (AA) | Artificial Analysis | 98.5% | – |
| Terminal-Bench 2.1 (Vals) | Vals AI | 80.5%† | 79.0% |
| MCP Atlas | Scale AI SEAL | 83.3%† | – |
| HiL-Bench | Scale AI SEAL | 56.3%† | – |
| τ³-bench Banking | τ²-bench | 39.7% | – |
| τ³-Banking (AA) | Artificial Analysis | 38.1% | 31.1% |
| ITBench SRE (AA) | Artificial Analysis | – | 40.3% |
| Analyst Agent (AA) | Artificial Analysis | 48.8% | – |
| APEX-Agents (AA) | Artificial Analysis | – | 35.8% |
| CyberBench | Vals AI | – | 73.6% |
| Finance Agent v2 | Vals AI | 56.3%† | 55.0% |
| Harvey's Legal Agent Benchmark v1 | Vals AI | 11.3%† | 1.3% |
| Legal Research Bench | Vals AI | 49.5%† | 36.5% |
| SkillsBench | Vals AI | – | 60.5% |
| SREBench | Vals AI | – | 1.5% |
| Tax Agent Bench | Vals AI | 38.3%† | 24.8% |
| Terminal-Bench 4.0 (Vals) | Vals AI | 41.4%† | 11.6% |
| Terminal-Bench Science | Vals AI | 15.7%† | 0.0% |
| Web Search Index | Vals AI | 46.9%† | – |
| Arena Search | Arena (LMArena) | 1229† | – |
| BALROG | BALROG | – | 45.6% |
| PostTrainBench | Epoch AI Benchmarking Hub | 41.8% | – |
| GBAEval | Epoch AI Benchmarking Hub | 74.5%† | – |
| Vending-Bench 2 | Andon Labs | 4966.6 | 4094.7† |
| AutomationBench | Artificial Analysis | 54.1% | 50.2% |
| GDP.pdf (AA) | Artificial Analysis | 24.0% | 24.0% |
| EnterpriseOps-Gym | Artificial Analysis | 51.1% | 40.8% |
| AA-Briefcase v1.1 | Artificial Analysis | 1537 | 1339 |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | – | 3.3% |
| FACTS Search | – | 79.3%† | |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 87.0% | 82.1% |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 90.2% | 61.0% |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 99.7% | 98.3% |
| ProofBench | Vals AI | 95.0% | 60.0% |
| LiveBench Mathematics | LiveBench | 96.0% | 87.2% |
| Arena Maths | Arena (LMArena) | 1521† | 1478† |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 70.7%† | 41.0% |
| LiveBench 1.0 Data Analysis | LiveBench | 80.5% | 78.0% |
| AA-Omniscience | Artificial Analysis | 43.3 | -10.3 |
| MMLU-Pro | Vals AI | 91.5%† | 86.0% |
| LegalBench | Vals AI | 88.6%† | 84.0% |
| CorpFin | Vals AI | 71.8%† | 64.2% |
| TaxEval | Vals AI | 76.9%† | 76.2% |
| PRBench Finance | Scale AI SEAL | 53.9%† | – |
| PRBench Legal | Scale AI SEAL | 52.6%† | – |
| BioMysteryBench | Vals AI | – | 61.5% |
| Excel Modeling Benchmark | Vals AI | 73.7%† | 67.1% |
| MedCode | Vals AI | 56.1%† | 42.4% |
| MedScribe | Vals AI | 88.5%† | 84.4% |
| Public Benefits Bench v1.1 | Vals AI | 70.4%† | 61.2% |
| EBR-bench | Epoch AI Benchmarking Hub | 39.5% | – |
| GDP.pdf | Epoch AI Benchmarking Hub | 29.8% | 22.7%† |
| AA-Omniscience: accuracy | Artificial Analysis | 65.3% | 42.7% |
| AA-Omniscience: non-hallucination | Artificial Analysis | 36.4% | 7.4% |
| Instruction following | |||
| LiveBench Language | LiveBench | 90.7% | 72.6% |
| LiveBench Instruction Following | LiveBench | 75.8% | 60.1% |
| IFBench | Artificial Analysis | 63.5% | – |
| Arena Instruction Following | Arena (LMArena) | 1508† | 1444† |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1503† | 1453† |
| EQ-Bench 4 | EQ-Bench | 1310† | 1149† |
| Arena Creative Writing | Arena (LMArena) | 1501† | 1410† |
| Arena Multi-turn | Arena (LMArena) | 1517† | 1459† |
| Arena Longer Queries | Arena (LMArena) | 1522† | 1452† |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | 1325† | 1257† |
| MMMU-Pro | Artificial Analysis | – | 78.5% |
| MMMU-Pro (Vals) | Vals AI | 89.3%† | 85.0% |
| MortgageTax | Vals AI | 68.9%† | 67.3% |
| SAGE | Vals AI | 51.9%† | 44.2% |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | 38.6%† | 22.6%† |
| Long context | |||
| AA-LCR | Artificial Analysis | 82.3% | 83.7% |
| Arena Document | Arena (LMArena) | 1499† | 1462† |
| MLCR | Artificial Analysis | 20.6% | 20.0% |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 163.4† | 156.5† |
| LiveBench 1.0 | LiveBench | 83.0% | 73.6% |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 49.6 | 37.3 |
| Vals RSI Index | Vals AI | 24.3% | – |
| Vals Science Index | Vals AI | – | 25.3% |
| Vals Multimodal Index | Vals AI | 74.2%† | 69.1% |
Data as of 2026-10-26. Best configuration of each model; every score links to its source on the model pages.
What changed
- GPT-5.6 Luna (max): 1 new result (AA-Briefcase v1.1).
- Claude Fable 5 (max): 1 new result (Vals RSI Index).
- GPT-5.6 Luna (max): 2 new results (Harvey's Legal Agent Benchmark v1, Vals Science Index).
- Claude Fable 5 (max): 1 score revision.
- GPT-5.6 Luna (max): 1 score revision.
- Claude Fable 5 (max): 1 new result (ForecastBench).
- GPT-5.6 Luna (max): BenchLeader Index 58.8 → 59.9.
- Claude Fable 5 (max): 4 new results (GDPval-AA v2.1, AA-Omniscience: accuracy, AA-Omniscience: non-hallucination…).
- GPT-5.6 Luna (max): 5 new results (AutomationBench, GDP.pdf (AA), MLCR…).
- GPT-5.6 Luna (max): 1 new result (SREBench); 1 score revision.
Related comparisons
- Claude Fable 5 vs Claude Opus 5.5
- Claude Fable 5 vs Gemini 4 Argon
- Claude Fable 5 vs Muse Wonder Full
- GPT-5.6 Luna vs Claude Opus 5.5
- GPT-5.6 Luna vs Gemini 4 Argon
- GPT-5.6 Luna vs Muse Wonder Full
- GPT-5.6 Luna vs Claude Fable 5.1
- Claude Fable 5 vs GPT-5.5
Rankings: Best LLM overall · Best LLM for coding · Best LLM for agents · Best cheap LLMs
Claude Fable 5 vs GPT-5.6 Luna: questions
- Is Claude Fable 5 better than GPT-5.6 Luna?
- Claude Fable 5 (max) leads on quality: 69.3 vs 59.9. The BenchLeader Index combines every independent quality benchmark; Claude Fable 5 (max) is ahead overall as of 2026-10-26, but check the category scores for your use.
- Which is cheaper, Claude Fable 5 or GPT-5.6 Luna, and will that change?
- GPT-5.6 Luna is cheaper today: $0.450 against $20.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is Claude Fable 5 better than GPT-5.6 Luna for coding?
- Claude Fable 5 scores higher in coding (74 vs 56 on the category index, where 50 is average). On SciCode, the top shared coding benchmark, it is 61.0% for Claude Fable 5 against 53.6% for GPT-5.6 Luna.
- Is Claude Fable 5 better than GPT-5.6 Luna for agents and tool use?
- Claude Fable 5 scores higher in agentic tasks (65 vs 49 on the category index, where 50 is average). On Terminal-Bench, the top shared agents & tools benchmark, it is 44.5% for Claude Fable 5 against 17.3% for GPT-5.6 Luna.
- Which is better for long documents?
- They are level on long-context benchmarks (66 vs 67). Both take about 1.1M tokens: Claude Fable 5 1,000,000, GPT-5.6 Luna 1,050,000.
- Which has lower latency?
- Claude Fable 5 (max) answers first: 101.8 s against 120.3 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
- Which is faster, Claude Fable 5 or GPT-5.6 Luna?
- GPT-5.6 Luna streams faster: 126 against 67 output tokens per second.
- Do independent benchmarks agree on Claude Fable 5 vs GPT-5.6 Luna?
- Across the 37 benchmarks both have, Claude Fable 5 leads 34, GPT-5.6 Luna leads 2 and 1 is a tie. Every overall index that rates both puts Claude Fable 5 ahead.
- When was GPT-5.6 Luna released?
- GPT-5.6 Luna was released on 9 Jul 2026.