Claude Sonnet 4.6 vs GPT-5.4 nano
Wins 6 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Long documents
Wins 2 of 8 areas
Math · Following instructions
Claude Sonnet 4.6 is the stronger all-rounder.GPT-5.4 nano is cheaper and better at following instructions.
Scores updated · 44 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.
- AgentsCarrying out multi-step tasks on its own70Claude Sonnet 4.67 of 7 tests
- CodingWriting and fixing software51Claude Sonnet 4.65 of 6 tests
- ReasoningHard problems that need careful thinking41Claude Sonnet 4.64 of 5 tests
- FactsGetting facts right instead of making them up31Claude Sonnet 4.63 of 4 tests
- Images and chartsUnderstanding pictures, charts and video20Claude Sonnet 4.62 of 2 tests
- Long documentsFinding answers in very long texts10Claude Sonnet 4.61 of 1 test
- Following instructionsDoing exactly what it is asked02GPT-5.4 nano2 of 2 tests
- MathCompetition and research-level math01GPT-5.4 nano1 of 1 test
Math and long documents 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.
GPT-5.4 nano costs 92% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Claude Sonnet 4.6 pulls ahead
- Abstract visual puzzles that people can solveARC-AGI-2+52.6points ahead
- Completes tasks by operating a computer desktopOSWorld-Verified+33.5points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+25.8points ahead
Where GPT-5.4 nano pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+19.3points ahead
- Research-level physics problemsCritPt+6.2points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+4.2points ahead
Every test, side by side
All 44 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Sonnet 4.6
- Terminal-Bench HardClaude Sonnet 4.6 by 10.65342.4+10.6
- Terminal-Bench 2.1Claude Sonnet 4.6 by 10.571.260.7+10.5
- LiveBench · CodingClaude Sonnet 4.6 by 8.579.370.8+8.5
- SWE-bench ProClaude Sonnet 4.6 by 5.758.152.4+5.7
- LiveBench · Agentic CodingGPT-5.4 nano by 4.242.646.8+4.2
- SciCodeClaude Sonnet 4.6 by 2.950.147.2+2.9
AgentsClaude Sonnet 4.6
- OSWorld-VerifiedClaude Sonnet 4.6 by 33.572.539+33.5
- AA IT-Bench SREClaude Sonnet 4.6 by 15.439.824.4+15.4
- GDPValClaude Sonnet 4.6 by 14.236.722.5+14.2
- τ-Bench V3 · BankingClaude Sonnet 4.6 by 734.427.4+7
- MCP AtlasClaude Sonnet 4.6 by 5.261.356.1+5.2
- AA ApexAgentsClaude Sonnet 4.6 by 3.12824.9+3.1
- Terminal-Bench 4.0Claude Sonnet 4.6 by 2.530.5+2.5
ReasoningClaude Sonnet 4.6
- ARC-AGI-2Claude Sonnet 4.6 by 52.658.35.7+52.6
- CritPtGPT-5.4 nano by 6.23.19.3+6.2
- GPQA DiamondClaude Sonnet 4.6 by 5.887.581.7+5.8
- Humanity's Last ExamClaude Sonnet 4.6 by 5.333.628.3+5.3
- LiveBench · ReasoningClaude Sonnet 4.6 by 3.784.881.1+3.7
FactsClaude Sonnet 4.6
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.6 by 25.851.625.8+25.8
- SimpleQA VerifiedClaude Sonnet 4.6 by 21.132.811.7+21.1
- AA-Omniscience · AccuracyClaude Sonnet 4.6 by 15.240.925.7+15.2
- Vectara HHEM hallucination ratelower is betterGPT-5.4 nano by 7.510.63.1+7.5
Images and chartsClaude Sonnet 4.6
- LMArena · VisionClaude Sonnet 4.6 by 87 rating points12831196+87 rating
- MMMU-ProClaude Sonnet 4.6 by 7.973.365.4+7.9
Following instructionsGPT-5.4 nano
- IFBenchGPT-5.4 nano by 19.356.675.9+19.3
- LiveBench · Instruction FollowingGPT-5.4 nano by 463.267.2+4
Other results16 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.
- AA-OmniscienceClaude Sonnet 4.6 by 41.712.2-29.5+41.7
- ARC-AGI-1Claude Sonnet 4.6 by 34.58651.5+34.5
- AA Agentic IndexClaude Sonnet 4.6 by 15.433.117.7+15.4
- livebench_languageClaude Sonnet 4.6 by 13.676.162.5+13.6
- Finance Agent v2Claude Sonnet 4.6 by 12.85138.2+12.8
- Terminal-Bench 2.0Claude Sonnet 4.6 by 12.859.146.3+12.8
- livebench_data_analysisClaude Sonnet 4.6 by 10.47867.6+10.4
- AA IntelligenceClaude Sonnet 4.6 by 9.430.120.7+9.4
- vectara_factual_consistencyGPT-5.4 nano by 7.589.496.9+7.5
- Artificial Analysis Coding IndexClaude Sonnet 4.6 by 6.96356.1+6.9
- ToolathlonClaude Sonnet 4.6 by 5.54135.5+5.5
- LiveBenchClaude Sonnet 4.6 by 5.475.570.1+5.4
- vectara_avg_summary_lengthGPT-5.4 nano by 29.7 rating points114.7144.4+29.7 rating
- frontiermath_tier_4_v1Claude Sonnet 4.6 by 28.36.3+2
- τ²-Bench Telecom (AA run)tie75.776tie
- vectara_answer_ratetie99.9100tie
Questions people ask
Which is better, Claude Sonnet 4.6 or GPT-5.4 nano?
Claude Sonnet 4.6 wins six of the eight areas we test: coding, agents, reasoning, facts, images and charts and long documents. GPT-5.4 nano wins math and following instructions, and costs 92% less.
Which is better for coding?
Claude Sonnet 4.6. It wins 5 of the 6 coding tests both models report; GPT-5.4 nano wins 1.
Which is cheaper?
Claude Sonnet 4.6 costs $3.00 per million input tokens and $15.00 per million output tokens; GPT-5.4 nano costs $0.20 and $1.25. That makes GPT-5.4 nano about 92% cheaper for the same work.
How do you compare the two?
We use the 44 benchmark tests both models have published scores on. The verdict counts the 28 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 16 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.