Claude Fable 5.1 vs GPT-6 Astra
Wins 4 of 8 areas
Coding · Agents · Images and charts · Long documents
Wins 4 of 8 areas
Reasoning · Facts · Math · Following instructions
The two are evenly matched.Claude Fable 5.1 is better at coding and agents; GPT-6 Astra at reasoning and math.
Both rank among the ten best models we track in reasoning, agents, facts and math.
Scores updated · 55 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 software41Claude Fable 5.14 of 5 tests
- AgentsCarrying out multi-step tasks on its own21Claude Fable 5.12 of 4 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video10Claude Fable 5.11 of 1 test
- Long documentsFinding answers in very long texts10Claude Fable 5.11 of 1 test
- ReasoningHard problems that need careful thinking24GPT-6 Astra4 of 6 tests
- MathCompetition and research-level math02GPT-6 Astra2 of 3 tests · 1 tie
- FactsGetting facts right instead of making them up12GPT-6 Astra2 of 3 tests
- Following instructionsDoing exactly what it is asked01GPT-6 Astra1 of 1 test
Images and charts, long documents and following instructions rest on a single test each.
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.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Claude Fable 5.1 pulls ahead
- Real work tasks from 44 professionsGDPVal+9.1points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+8.8points ahead
- Code for real scientific research problemsSciCode+6.6points ahead
Where GPT-6 Astra pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+21.3points ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+9.8points ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+7.1points ahead
Every test, side by side
All 55 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Fable 5.1
- LiveBench · Agentic CodingClaude Fable 5.1 by 8.866.157.3+8.8
- SciCodeClaude Fable 5.1 by 6.663.156.5+6.6
- LiveBench · CodingClaude Fable 5.1 by 686.480.4+6
- LMArena · WebDevGPT-6 Astra by 32 rating points17431775+32 rating
- Terminal-Bench 2.1Claude Fable 5.1 by 391.488.4+3
AgentsClaude Fable 5.1
- GDPValClaude Fable 5.1 by 9.162.753.6+9.1
- Terminal-Bench 4.0GPT-6 Astra by 7.15259.1+7.1
- τ-Bench V3 · BankingClaude Fable 5.1 by 5.847.241.4+5.8
- AA IT-Bench SREtie49.548.6tie
ReasoningGPT-6 Astra
- ARC-AGI-2GPT-6 Astra by 59095+5
- Humanity's Last ExamClaude Fable 5.1 by 4.459.154.7+4.4
- SimpleBenchClaude Fable 5.1 by 386.683.6+3
- GPQA DiamondGPT-6 Astra by 2.493.796.1+2.4
- CritPtGPT-6 Astra by 229.731.7+2
- LiveBench · ReasoningGPT-6 Astra by 191.792.7+1
FactsGPT-6 Astra
- AA-Omniscience · Non-hallucinationGPT-6 Astra by 21.327.448.7+21.3
- SimpleQA VerifiedGPT-6 Astra by 4.870.875.6+4.8
- AA-Omniscience · AccuracyClaude Fable 5.1 by 4.667.262.6+4.6
Images and chartsClaude Fable 5.1
- LMArena · VisionClaude Fable 5.1 by 39 rating points13191280+39 rating
MathGPT-6 Astra
- FrontierMath Tier 4GPT-6 Astra by 9.887.897.6+9.8
- FrontierMath Tiers 1-3 (v2)GPT-6 Astra by 3.590.293.7+3.5
- LiveBench · Mathematicstie9796.8tie
Following instructionsGPT-6 Astra
- LiveBench · Instruction FollowingGPT-6 Astra by 2.67375.6+2.6
Other results31 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 2.1Claude Fable 5.1 by 193 rating points17351542+193 rating
- Science and math (Terminal-Bench Science 0.1)GPT-6 Astra by 15.552.668.1+15.5
- Terminal-Bench-Science 0.1GPT-6 Astra by 1252.664.6+12
- AutomationBench (Score)GPT-6 Astra by 1031.441.4+10
- Cybersecurity (CWE-bench v1)GPT-6 Astra by 105868+10
- Knowledge work (AutomationBench)GPT-6 Astra by 1031.441.4+10
- HLE (with tools)Claude Fable 5.1 by 8.465.657.2+8.4
- LVBenchGPT-6 Astra by 7.879.787.5+7.8
- Multimodal understanding (LVBench)GPT-6 Astra by 7.879.787.5+7.8
- GraphWalks (Up to 128k, BFS (F1))GPT-6 Astra by 7.391.498.7+7.3
- GraphWalks (256k to 1M, BFS (F1))GPT-6 Astra by 6.86571.8+6.8
- DeepSWE 1.1GPT-6 Astra by 6.767.474.1+6.7
- AA Agentic IndexClaude Fable 5.1 by 6.55851.5+6.5
- RiemannBenchGPT-6 Astra by 6.465.672+6.4
- Science and math (RiemannBench)GPT-6 Astra by 6.465.672+6.4
- OSWorld 2.0Claude Fable 5.1 by 5.377.972.6+5.3
- Artificial Analysis Coding IndexClaude Fable 5.1 by 4.781.676.9+4.7
- PostTrainBenchGPT-6 Astra by 4.140.244.3+4.1
- HiL-Bench (Tools-allowed)GPT-6 Astra by 3.861.565.3+3.8
- Knowledge work (Vals Index)Claude Fable 5.1 by 2.765.863.1+2.7
- livebench_data_analysisGPT-6 Astra by 2.780.383+2.7
- Vals IndexClaude Fable 5.1 by 2.765.863.1+2.7
- HealthBench Professional length-adjustedGPT-6 Astra by 2.662.164.7+2.6
- Terminal-Bench 4.0GPT-6 Astra by 1.955.857.7+1.9
- HealthBench length-adjustedClaude Fable 5.1 by 1.76058.3+1.7
- HealthBench ProfessionalGPT-6 Astra by 1.362.163.4+1.3
- AA Intelligencetie53.452.7tie
- Agentic coding (Terminal-bench 4.0)tie57.958.2tie
- AA-Omnisciencetie43.543.4tie
- livebench_languagetie89.589.4tie
- ARC-AGI-1tie97.597.5tie
Questions people ask
Which is better, Claude Fable 5.1 or GPT-6 Astra?
Claude Fable 5.1 and GPT-6 Astra each win four of the eight areas we test. Claude Fable 5.1 wins coding, agents, images and charts and long documents; GPT-6 Astra wins reasoning, facts, math and following instructions.
Which is better for coding?
Claude Fable 5.1. It wins 4 of the 5 coding tests both models report; GPT-6 Astra wins 1.
Which is cheaper?
Claude Fable 5.1 costs $10.00 per million input tokens and $50.00 per million output tokens; GPT-6 Astra costs $10.00 and $50.00. For a million tokens read plus a million written, they cost about the same.
How do you compare the two?
We use the 55 benchmark tests both models have published scores on. The verdict counts the 24 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 31 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.