Claude Fable 5 vs GPT-5.5
Wins 6 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math
Wins 1 of 8 areas
Long documents
Claude Fable 5 is the stronger all-rounder.GPT-5.5 is cheaper and better at long documents.
Both rank among the ten best models we track in reasoning, coding, agents and math.
Scores updated · 63 tests both models report · How we compare
Where each one wins
Tests won in each of the eight areas we test. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- CodingWriting and fixing software60Claude Fable 56 of 7 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own60Claude Fable 56 of 7 tests · 1 tie
- ReasoningHard problems that need careful thinking40Claude Fable 54 of 6 tests · 2 ties
- FactsGetting facts right instead of making them up30Claude Fable 53 of 3 tests
- MathCompetition and research-level math20Claude Fable 52 of 3 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video20Claude Fable 52 of 2 tests
- Long documentsFinding answers in very long texts01GPT-5.51 of 1 test
- Following instructionsDoing exactly what it is asked11Even1 each
Long documents rests on a single test.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
GPT-5.5 costs 42% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Claude Fable 5 pulls ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+27.8points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+25.4points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+21.4points ahead
Every test, side by side
All 63 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Fable 5
- SWE-bench ProClaude Fable 5 by 21.48058.6+21.4
- LMArena · WebDevClaude Fable 5 by 111 rating points16241513+111 rating
- LiveBench · Agentic CodingClaude Fable 5 by 8.262.254+8.2
- SciCodeClaude Fable 5 by 5.26155.8+5.2
- LiveBench · CodingClaude Fable 5 by 3.88682.2+3.8
- Terminal-Bench HardClaude Fable 5 by 2.362.960.6+2.3
- Terminal-Bench 2.1tie84.684.3tie
AgentsClaude Fable 5
- Terminal-Bench 4.0Claude Fable 5 by 27.842.414.6+27.8
- GDPValClaude Fable 5 by 12.955.642.7+12.9
- MCP AtlasClaude Fable 5 by 9.484.775.3+9.4
- OSWorld-VerifiedClaude Fable 5 by 6.38578.7+6.3
- AA ApexAgentsClaude Fable 5 by 5.643.337.7+5.6
- BrowseCompClaude Fable 5 by 387.484.4+3
- τ-Bench V3 · Bankingtie38.139tie
ReasoningClaude Fable 5
- SimpleBenchClaude Fable 5 by 12.981.969+12.9
- Humanity's Last ExamClaude Fable 5 by 9.755.545.8+9.7
- ARC-AGI-2Claude Fable 5 by 4.289.285+4.2
- CritPtClaude Fable 5 by 1.528.627.1+1.5
- GPQA Diamondtie92.693.5tie
- LiveBench · Reasoningtie89.789.7tie
FactsClaude Fable 5
- AA-Omniscience · Non-hallucinationClaude Fable 5 by 25.436.411+25.4
- SimpleQA VerifiedClaude Fable 5 by 7.770.763+7.7
- AA-Omniscience · AccuracyClaude Fable 5 by 7.465.358+7.4
Images and chartsClaude Fable 5
- MMMU-ProClaude Fable 5 by 4.384.279.9+4.3
- LMArena · VisionClaude Fable 5 by 33 rating points13251292+33 rating
MathClaude Fable 5
- FrontierMath Tier 4Claude Fable 5 by 17.790.272.5+17.7
- FrontierMath Tiers 1-3 (v2)Claude Fable 5 by 1.78785.3+1.7
- LiveBench · Mathematicstie9695.9tie
Following instructionsEven
- IFBenchGPT-5.5 by 12.463.575.9+12.4
- LiveBench · Instruction FollowingClaude Fable 5 by 5.175.870.7+5.1
Other results32 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- GDPval-AA v2 EloClaude Fable 5 by 232 rating points17231491+232 rating
- SWE-MarathonClaude Fable 5 by 233512+23
- AA-OmniscienceClaude Fable 5 by 22.843.320.5+22.8
- GDPval-AA v2Claude Fable 5 by 214 rating points17231509+214 rating
- JobBenchClaude Fable 5 by 19.157.438.3+19.1
- HiL-Bench (Tools-allowed)Claude Fable 5 by 16.656.339.7+16.6
- GDPval-AA (Elo)Claude Fable 5 by 163 rating points19321769+163 rating
- Program BenchClaude Fable 5 by 15.586.370.8+15.5
- MLS Bench Litelower is betterGPT-5.5 by 14.449.935.5+14.4
- HealthBench ProfessionalClaude Fable 5 by 14.26651.8+14.2
- AA Agentic IndexClaude Fable 5 by 13.75137.3+13.7
- PostTrainBenchClaude Fable 5 by 1341.428.4+13
- HLE (with tools)Claude Fable 5 by 11.663.852.2+11.6
- AA IntelligenceClaude Fable 5 by 11.249.638.4+11.2
- Legal Agent BenchmarkClaude Fable 5 by 11.213.32.1+11.2
- Legal Agent Benchmark (Harvey's Held-Out Set)Claude Fable 5 by 11.213.32.1+11.2
- ToolathlonClaude Fable 5 by 6.161.755.6+6.1
- SpreadsheetBench 2Claude Fable 5 by 5.634.729.1+5.6
- MCP Mark VerifiedGPT-5.5 by 5.587.492.9+5.5
- GDP (Surge AI)Claude Fable 5 by 4.929.824.9+4.9
- τ²-Bench Telecom (AA run)Claude Fable 5 by 4.698.593.9+4.6
- Finance Agent v2Claude Fable 5 by 4.556.351.8+4.5
- Toolathlon VerifiedClaude Fable 5 by 4.477.973.5+4.4
- OfficeQA ProClaude Fable 5 by 3.857.954.1+3.8
- ARC-AGI-1Claude Fable 5 by 3.598.595+3.5
- livebench_languageClaude Fable 5 by 3.390.787.4+3.3
- DeepSWE 1.1Claude Fable 5 by 37067+3
- LiveBenchGPT-5.5 by 2.478.380.7+2.4
- Artificial Analysis Coding IndexClaude Fable 5 by 1.676.574.9+1.6
- livebench_data_analysisGPT-5.5 by 1.180.581.6+1.1
- Agents' Last Examtie25.726.6tie
- DeepSWEtie7070tie
Questions people ask
Which is better, Claude Fable 5 or GPT-5.5?
Claude Fable 5 wins six of the eight areas we test: coding, agents, reasoning, facts, images and charts and math. GPT-5.5 wins long documents, and costs 42% less. They are level on following instructions.
Which is better for coding?
Claude Fable 5. It wins 6 of the 7 coding tests both models report; GPT-5.5 wins none, and 1 is a tie.
Which is cheaper?
Claude Fable 5 costs $10.00 per million input tokens and $50.00 per million output tokens; GPT-5.5 costs $5.00 and $30.00. That makes GPT-5.5 about 42% cheaper for the same work.
How do you compare the two?
We use the 63 benchmark tests both models have published scores on. The verdict counts the 31 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 32 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.