Claude Sonnet 4.5 vs GPT-5
Wins 1 of 8 areas
Coding
Wins 5 of 8 areas
Reasoning · Images and charts · Math · Long documents · Following instructions
GPT-5 is the stronger all-rounder.Claude Sonnet 4.5 is better at coding.
Scores updated · 82 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.
- ReasoningHard problems that need careful thinking14GPT-54 of 5 tests
- Images and chartsUnderstanding pictures, charts and video03GPT-53 of 3 tests
- MathCompetition and research-level math03GPT-53 of 3 tests
- Following instructionsDoing exactly what it is asked02GPT-52 of 2 tests
- Long documentsFinding answers in very long texts01GPT-51 of 1 test
- CodingWriting and fixing software62Claude Sonnet 4.56 of 8 tests
- AgentsCarrying out multi-step tasks on its own11Even1 each · 1 tie
- FactsGetting facts right instead of making them up22Even2 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 costs 38% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where GPT-5 pulls ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+31.5points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+30.8points ahead
- Recent programming contest problemsLiveCodeBench v6+23points ahead
Where Claude Sonnet 4.5 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+33.1points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+20.6points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+11.7points ahead
Every test, side by side
All 82 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Sonnet 4.5
- LiveCodeBench v6GPT-5 by 236487+23
- Terminal-Bench 2.1Claude Sonnet 4.5 by 20.655.835.2+20.6
- SWE-bench MultilingualClaude Sonnet 4.5 by 11.76755.3+11.7
- Terminal-Bench HardClaude Sonnet 4.5 by 335.632.6+3
- SciCodeClaude Sonnet 4.5 by 2.845.742.9+2.8
- LMArena · WebDevGPT-5 by 24 rating points13941418+24 rating
- SWE-bench VerifiedClaude Sonnet 4.5 by 2.377.274.9+2.3
- SWE-bench ProClaude Sonnet 4.5 by 1.843.641.8+1.8
AgentsEven
- BrowseCompGPT-5 by 30.824.154.9+30.8
- τ-Bench V3 · BankingClaude Sonnet 4.5 by 2.424.522.1+2.4
- GDPValtie20.521tie
ReasoningGPT-5
- Humanity's Last ExamGPT-5 by 10.717.828.5+10.7
- CritPtGPT-5 by 4.61.15.7+4.6
- ARC-AGI-2Claude Sonnet 4.5 by 3.713.69.9+3.7
- SimpleBenchGPT-5 by 2.454.356.7+2.4
- GPQA DiamondGPT-5 by 283.485.4+2
FactsEven
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.5 by 33.150.917.8+33.1
- SimpleQA VerifiedGPT-5 by 19.430.750.1+19.4
- AA-Omniscience · AccuracyGPT-5 by 7.432.940.3+7.4
- Vectara HHEM hallucination ratelower is betterClaude Sonnet 4.5 by 3.11215.1+3.1
Images and chartsGPT-5
- CharXiv (RQ)GPT-5 by 13.967.281.1+13.9
- MMMU-ProGPT-5 by 5.568.774.2+5.5
- MMMUGPT-5 by 4.679.684.2+4.6
MathGPT-5
- FrontierMath Tiers 1-3 (v2)GPT-5 by 31.523.955.4+31.5
- FrontierMath Tier 4GPT-5 by 19.62.421.9+19.6
- IMOAnswerBenchGPT-5 by 10.165.976+10.1
Following instructionsGPT-5
- IFBenchGPT-5 by 15.857.373.1+15.8
- Multi-ChallengeGPT-5 by 14.355.369.6+14.3
Other results53 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.
- BrowseComp w/ toolsGPT-5 by 30.824.154.9+30.8
- OJ-Bench (cpp)GPT-5 by 25.830.456.2+25.8
- OJ-Bench (cpp) no toolsGPT-5 by 25.830.456.2+25.8
- HealthBenchGPT-5 by 2344.267.2+23
- HealthBench no toolsGPT-5 by 2344.267.2+23
- LiveCodeBenchV6 no toolsGPT-5 by 236487+23
- HMMT Feb. 2025GPT-5 by 20.867.588.3+20.8
- BrowseComp-ZHGPT-5 by 20.642.463+20.6
- BrowseComp-ZH w/ toolsGPT-5 by 20.642.463+20.6
- HMMT 2025GPT-5 by 18.774.693.3+18.7
- HMMT25 no toolsGPT-5 by 18.774.693.3+18.7
- τ²-BenchClaude Sonnet 4.5 by 15.69882.4+15.6
- Arena-Hard (Creative Writing)GPT-5 by 15.576.792.2+15.5
- Terminal-Bench 2.0Claude Sonnet 4.5 by 14.85035.2+14.8
- Artificial Analysis Coding IndexClaude Sonnet 4.5 by 14.352.137.8+14.3
- LiveCodeBenchGPT-5 by 12.57183.5+12.5
- xbench-DeepSearchGPT-5 by 11.86677.8+11.8
- ArtifactsBenchGPT-5 by 11.561.573+11.5
- AA Agentic IndexGPT-5 by 917.526.5+9
- AA-OmniscienceClaude Sonnet 4.5 by 8.6-0.1-8.7+8.6
- Arena-Hard (Hard Prompt)GPT-5 by 8.663.371.9+8.6
- Longform Writing eval (Kimi K2 Thinking system card)Claude Sonnet 4.5 by 8.479.871.4+8.4
- frontiermath_tier_4_v1GPT-5 by 8.34.212.5+8.3
- AIME 2025GPT-5 by 7.68794.6+7.6
- HMMT Nov. 2025GPT-5 by 7.581.789.2+7.5
- Terminal-Bench w/ simulated tools (JSON)Claude Sonnet 4.5 by 7.25143.8+7.2
- τ²-Bench Telecom (AA run)GPT-5 by 6.778.184.8+6.7
- AIME25 no toolsGPT-5 by 6.68894.6+6.6
- swe_bench_bashClaude Sonnet 4.5 by 6.471.465+6.4
- Terminal-BenchClaude Sonnet 4.5 by 6.25043.8+6.2
- HarmBenchGPT-5 by 5.69297.6+5.6
- GAIA (text only)GPT-5 by 5.271.276.4+5.2
- τ²-Bench (Retail)Claude Sonnet 4.5 by 5.186.281.1+5.1
- FinSearchComp-T3 (Tools-allowed)GPT-5 by 4.54448.5+4.5
- FinSearchComp-T3 w/ toolsGPT-5 by 4.54448.5+4.5
- vectara_answer_rateGPT-5 by 4.395.699.9+4.3
- vectara_avg_summary_lengthGPT-5 by 34.9 rating points127.8162.7+34.9 rating
- HLE (with tools)GPT-5 by 3.23235.2+3.2
- FinSearchComp-globalGPT-5 by 3.160.863.9+3.1
- vectara_factual_consistencyClaude Sonnet 4.5 by 3.18884.9+3.1
- matharena_visual_math_overallGPT-5 by 375.878.8+3
- AA IntelligenceGPT-5 by 2.320.723+2.3
- GPQA (unspecified)GPT-5 by 2.383.485.7+2.3
- AIR-Bench 2024Claude Sonnet 4.5 by 2.189.887.7+2.1
- bbqClaude Sonnet 4.5 by 2.198.996.8+2.1
- ARC-AGI-1GPT-5 by 263.765.7+2
- Seal-0 w/ toolsClaude Sonnet 4.5 by 253.451.4+2
- Frames w/ toolsGPT-5 by 18586+1
- anthropic_red_teamtie98.299.1tie
- XSTesttie96.497.1tie
- MMLU-Protie87.587.1tie
- MMLU-Reduxtie95.695.3tie
- simple_safety_teststie10099.8tie
Questions people ask
Which is better, Claude Sonnet 4.5 or GPT-5?
GPT-5 wins five of the eight areas we test: reasoning, images and charts, math, long documents and following instructions. Claude Sonnet 4.5 wins coding. They are level on agents and facts.
Which is better for coding?
Claude Sonnet 4.5. It wins 6 of the 8 coding tests both models report; GPT-5 wins 2.
Which is cheaper?
Claude Sonnet 4.5 costs $3.00 per million input tokens and $15.00 per million output tokens; GPT-5 costs $1.25 and $10.00. That makes GPT-5 about 38% cheaper for the same work.
How do you compare the two?
We use the 82 benchmark tests both models have published scores on. The verdict counts the 29 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 53 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.