Claude Opus 4.6 vs Claude Sonnet 4.5
Wins 7 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Long documents
Wins 1 of 8 areas
Following instructions
Claude Opus 4.6 is the stronger all-rounder.Claude Sonnet 4.5 is cheaper and better at following instructions.
Scores updated · 54 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 software70Claude Opus 4.67 of 7 tests
- ReasoningHard problems that need careful thinking50Claude Opus 4.65 of 5 tests
- AgentsCarrying out multi-step tasks on its own40Claude Opus 4.64 of 4 tests
- MathCompetition and research-level math30Claude Opus 4.63 of 3 tests
- Images and chartsUnderstanding pictures, charts and video20Claude Opus 4.62 of 2 tests
- FactsGetting facts right instead of making them up21Claude Opus 4.62 of 4 tests · 1 tie
- Long documentsFinding answers in very long texts10Claude Opus 4.61 of 1 test
- Following instructionsDoing exactly what it is asked02Claude Sonnet 4.52 of 2 tests
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.
Claude Sonnet 4.5 costs 40% 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 Opus 4.6 pulls ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+59.9points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+55.2points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+42.1points ahead
Where Claude Sonnet 4.5 pulls ahead
- Keeps track of context across a multi-turn chatMulti-Challenge+18.1points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+13.7points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+4.2points ahead
Every test, side by side
All 54 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 4.6
- LiveCodeBench v6Claude Opus 4.6 by 24.888.864+24.8
- LMArena · WebDevClaude Opus 4.6 by 152 rating points15461394+152 rating
- SWE-bench MultilingualClaude Opus 4.6 by 10.877.867+10.8
- Terminal-Bench HardClaude Opus 4.6 by 10.646.235.6+10.6
- SWE-bench ProClaude Opus 4.6 by 9.853.443.6+9.8
- SciCodeClaude Opus 4.6 by 6.251.945.7+6.2
- SWE-bench VerifiedClaude Opus 4.6 by 3.680.877.2+3.6
AgentsClaude Opus 4.6
- BrowseCompClaude Opus 4.6 by 59.98424.1+59.9
- GDPValClaude Opus 4.6 by 35.455.920.5+35.4
- MCP AtlasClaude Opus 4.6 by 18.962.743.8+18.9
- OSWorld-VerifiedClaude Opus 4.6 by 11.372.761.4+11.3
ReasoningClaude Opus 4.6
- ARC-AGI-2Claude Opus 4.6 by 55.268.813.6+55.2
- Humanity's Last ExamClaude Opus 4.6 by 22.139.917.8+22.1
- SimpleBenchClaude Opus 4.6 by 13.367.654.3+13.3
- CritPtClaude Opus 4.6 by 11.512.61.1+11.5
- GPQA DiamondClaude Opus 4.6 by 6.289.683.4+6.2
FactsClaude Opus 4.6
- SimpleQA VerifiedClaude Opus 4.6 by 16.34730.7+16.3
- AA-Omniscience · AccuracyClaude Opus 4.6 by 14.14732.9+14.1
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.5 by 13.737.250.9+13.7
- Vectara HHEM hallucination ratelower is bettertie12.212tie
Images and chartsClaude Opus 4.6
- CharXiv (RQ)Claude Opus 4.6 by 10.277.467.2+10.2
- MMMU-ProClaude Opus 4.6 by 6.775.468.7+6.7
MathClaude Opus 4.6
- FrontierMath Tiers 1-3 (v2)Claude Opus 4.6 by 42.16623.9+42.1
- FrontierMath Tier 4Claude Opus 4.6 by 24.426.82.4+24.4
- IMOAnswerBenchClaude Opus 4.6 by 9.475.365.9+9.4
Following instructionsClaude Sonnet 4.5
- Multi-ChallengeClaude Sonnet 4.5 by 18.137.155.3+18.1
- IFBenchClaude Sonnet 4.5 by 4.253.157.3+4.2
Other results26 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.
- MHGRN (8-needle 1M variant)Claude Opus 4.6 by 57.57618.5+57.5
- Needle-in-a-Haystack (8-needle 1M variant)Claude Opus 4.6 by 57.57618.5+57.5
- AA Agentic IndexClaude Opus 4.6 by 50.167.617.5+50.1
- ARC-AGI-1Claude Opus 4.6 by 29.39363.7+29.3
- HLE (with tools)Claude Opus 4.6 by 21.353.332+21.3
- frontiermath_tier_4_v1Claude Opus 4.6 by 18.722.94.2+18.7
- Toolathlon Pass@∞Claude Opus 4.6 by 18.547.228.7+18.5
- LiveCodeBenchClaude Opus 4.6 by 17.888.871+17.8
- ToolathlonClaude Opus 4.6 by 15.856.841+15.8
- Terminal-Bench 2.0Claude Opus 4.6 by 15.465.450+15.4
- Toolathlon Avg turnsClaude Sonnet 4.5 by 15.116.932+15.1
- HMMT Nov. 2025Claude Opus 4.6 by 14.696.381.7+14.6
- τ²-Bench Telecom (AA run)Claude Opus 4.6 by 1492.178.1+14
- AA-OmniscienceClaude Opus 4.6 by 13.813.7-0.1+13.8
- AIME 2025Claude Opus 4.6 by 12.899.887+12.8
- AA IntelligenceClaude Opus 4.6 by 11.231.920.7+11.2
- AIME25 no toolsClaude Opus 4.6 by 7.695.688+7.6
- τ²-Bench (Retail)Claude Opus 4.6 by 5.791.986.2+5.7
- vectara_answer_rateClaude Opus 4.6 by 4.299.895.6+4.2
- Artificial Analysis Coding IndexClaude Sonnet 4.5 by 448.152.1+4
- RealWorldQAClaude Opus 4.6 by 3.673.970.3+3.6
- matharena_visual_math_overallClaude Sonnet 4.5 by 3.572.375.8+3.5
- MMMLUClaude Opus 4.6 by 291.189.1+2
- MMLU-ProClaude Opus 4.6 by 1.689.187.5+1.6
- vectara_avg_summary_lengthtie137.6127.8tie
- vectara_factual_consistencytie87.888tie
Questions people ask
Which is better, Claude Opus 4.6 or Claude Sonnet 4.5?
Claude Opus 4.6 wins seven of the eight areas we test: coding, agents, reasoning, facts, images and charts, math and long documents. Claude Sonnet 4.5 wins following instructions, and costs 40% less.
Which is better for coding?
Claude Opus 4.6. It wins 7 of the 7 coding tests both models report; Claude Sonnet 4.5 wins none.
Which is cheaper?
Claude Opus 4.6 costs $5.00 per million input tokens and $25.00 per million output tokens; Claude Sonnet 4.5 costs $3.00 and $15.00. That makes Claude Sonnet 4.5 about 40% cheaper for the same work.
How do you compare the two?
We use the 54 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 26 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.