Claude Sonnet 4 vs Claude Opus 5
Wins 0 of 6 areas
—
Wins 5 of 6 areas
Coding · Agents · Reasoning · Images and charts · Long documents
Claude Opus 5 is the stronger all-rounder.Claude Sonnet 4 is cheaper.
Scores updated · 25 tests both models report · How we compare
Where each one wins
Tests won in each of the six areas where both have results. 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 software05Claude Opus 55 of 5 tests
- ReasoningHard problems that need careful thinking05Claude Opus 55 of 5 tests
- AgentsCarrying out multi-step tasks on its own04Claude Opus 54 of 4 tests
- Images and chartsUnderstanding pictures, charts and video02Claude Opus 52 of 2 tests
- Long documentsFinding answers in very long texts01Claude Opus 51 of 1 test
- FactsGetting facts right instead of making them up11Even1 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.
Claude Sonnet 4 costs 40% 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 Opus 5 pulls ahead
- Abstract visual puzzles that people can solveARC-AGI-2+84.5points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+78.6points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+52.8points ahead
Where Claude Sonnet 4 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+31.7points ahead
Every test, side by side
All 25 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 5
- Terminal-Bench 2.1Claude Opus 5 by 52.836.389.1+52.8
- SWE-bench ProClaude Opus 5 by 36.542.779.2+36.5
- SWE-bench MultilingualClaude Opus 5 by 32.656.989.5+32.6
- SWE-bench VerifiedClaude Opus 5 by 23.372.796+23.3
- SciCodeClaude Opus 5 by 16.44056.4+16.4
AgentsClaude Opus 5
- BrowseCompClaude Opus 5 by 78.612.290.8+78.6
- GDPValClaude Opus 5 by 51.59.761.2+51.5
- OSWorld-VerifiedClaude Opus 5 by 41.242.283.4+41.2
- τ-Bench V3 · BankingClaude Opus 5 by 25.416.742.1+25.4
ReasoningClaude Opus 5
- ARC-AGI-2Claude Opus 5 by 84.55.990.4+84.5
- Humanity's Last ExamClaude Opus 5 by 44.210.754.9+44.2
- SimpleBenchClaude Opus 5 by 35.145.580.6+35.1
- CritPtClaude Opus 5 by 28.80.329.1+28.8
- GPQA DiamondClaude Opus 5 by 15.577.793.2+15.5
FactsEven
- AA-Omniscience · AccuracyClaude Opus 5 by 38.222.760.9+38.2
- AA-Omniscience · Non-hallucinationClaude Sonnet 4 by 31.770.939.2+31.7
Images and chartsClaude Opus 5
- MMMU-ProClaude Opus 5 by 22.961.884.7+22.9
- LMArena · VisionClaude Opus 5 by 130 rating points11911321+130 rating
Other results6 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.
- ARC-AGI-1Claude Opus 5 by 57.54097.5+57.5
- HLE (with tools)Claude Opus 5 by 43.320.363.6+43.3
- Artificial Analysis Coding IndexClaude Opus 5 by 40.437.678+40.4
- AA Agentic IndexClaude Opus 5 by 38.617.656.2+38.6
- AA-OmniscienceClaude Opus 5 by 36.90.237.1+36.9
- AA IntelligenceClaude Opus 5 by 31.918.950.8+31.9
Questions people ask
Which is better, Claude Sonnet 4 or Claude Opus 5?
Claude Opus 5 wins five of the six areas where both have results: coding, agents, reasoning, images and charts and long documents. Claude Sonnet 4 wins none, but costs 40% less. They are level on facts.
Which is better for coding?
Claude Opus 5. It wins 5 of the 5 coding tests both models report; Claude Sonnet 4 wins none.
Which is cheaper?
Claude Sonnet 4 costs $3.00 per million input tokens and $15.00 per million output tokens; Claude Opus 5 costs $5.00 and $25.00. That makes Claude Sonnet 4 about 40% cheaper for the same work.
How do you compare the two?
We use the 25 benchmark tests both models have published scores on. The verdict counts the 19 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 6 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.