Claude Opus 4.8 vs Sonnet 5.5
Wins 2 of 6 areas
Facts · Following instructions
Wins 3 of 6 areas
Coding · Reasoning · Math
Sonnet 5.5 wins more areas, narrowly.Claude Opus 4.8 is better at facts.
Scores updated · 28 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 software05Sonnet 5.55 of 5 tests
- ReasoningHard problems that need careful thinking03Sonnet 5.53 of 3 tests
- MathCompetition and research-level math03Sonnet 5.53 of 3 tests
- FactsGetting facts right instead of making them up10Claude Opus 4.81 of 1 test
- Following instructionsDoing exactly what it is asked10Claude Opus 4.81 of 1 test
- Images and chartsUnderstanding pictures, charts and video00Even0 each · 1 tie
Facts, images and charts 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.
Sonnet 5.5 costs 60% 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 Sonnet 5.5 pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+24.4points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+15.8points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+12.1points ahead
Where Claude Opus 4.8 pulls ahead
- Follows detailed instructions exactlyLiveBench · Instruction Following+15.2points ahead
- Short factual questions, answered correctlySimpleQA Verified+6.5points ahead
Every test, side by side
All 28 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingSonnet 5.5
- LMArena · WebDevSonnet 5.5 by 230 rating points15561786+230 rating
- SWE-bench ProSonnet 5.5 by 12.169.281.3+12.1
- LiveBench · CodingSonnet 5.5 by 9.681.891.4+9.6
- SWE-bench MultilingualSonnet 5.5 by 5.984.490.3+5.9
- LiveBench · Agentic CodingSonnet 5.5 by 5.850.556.3+5.8
ReasoningSonnet 5.5
- Humanity's Last ExamSonnet 5.5 by 15.848.764.5+15.8
- SimpleBenchSonnet 5.5 by 11.164.875.9+11.1
- LiveBench · ReasoningSonnet 5.5 by 2.489.291.6+2.4
MathSonnet 5.5
- FrontierMath Tier 4Sonnet 5.5 by 24.456.180.5+24.4
- FrontierMath Tiers 1-3 (v2)Sonnet 5.5 by 8.88088.8+8.8
- LiveBench · MathematicsSonnet 5.5 by 1.894.396.1+1.8
Following instructionsClaude Opus 4.8
- LiveBench · Instruction FollowingClaude Opus 4.8 by 15.27256.8+15.2
Other results14 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.
- Terminal-Bench 4.0Sonnet 5.5 by 4723.670.6+47
- SWE-bench MultimodalSonnet 5.5 by 15.938.454.3+15.9
- AA IntelligenceSonnet 5.5 by 14.241.856+14.2
- HealthBench ProfessionalSonnet 5.5 by 13.455.869.2+13.4
- DeepSWE 1.1Sonnet 5.5 by 125971+12
- Program BenchSonnet 5.5 by 7.871.979.7+7.8
- HLE (with tools)Sonnet 5.5 by 6.657.964.5+6.6
- livebench_data_analysisClaude Opus 4.8 by 6.56659.5+6.5
- HealthBenchSonnet 5.5 by 6.159.365.4+6.1
- Toolathlon VerifiedClaude Opus 4.8 by 2.179.977.8+2.1
- Legal Agent BenchmarkSonnet 5.5 by 1.71011.7+1.7
- livebench_languageClaude Opus 4.8 by 1.779.778+1.7
- OfficeQAtie77.676.9tie
- OfficeQA Protie66.265.6tie
Questions people ask
Which is better, Claude Opus 4.8 or Sonnet 5.5?
Sonnet 5.5 wins three of the six areas where both have results: coding, reasoning and math. Claude Opus 4.8 wins facts and following instructions. They are level on images and charts.
Which is better for coding?
Sonnet 5.5. It wins 5 of the 5 coding tests both models report; Claude Opus 4.8 wins none.
Which is cheaper?
Claude Opus 4.8 costs $5.00 per million input tokens and $25.00 per million output tokens; Sonnet 5.5 costs $2.00 and $10.00. That makes Sonnet 5.5 about 60% cheaper for the same work.
How do you compare the two?
We use the 28 benchmark tests both models have published scores on. The verdict counts the 14 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 14 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.