Claude Opus 4.5 vs Qwen3.5 27B
Wins 5 of 8 areas
Coding · Reasoning · Facts · Math · Long documents
Wins 2 of 8 areas
Images and charts · Following instructions
Claude Opus 4.5 is the stronger all-rounder.Qwen3.5 27B is cheaper and better at images and charts.
Scores updated · 57 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 software51Claude Opus 4.55 of 7 tests · 1 tie
- FactsGetting facts right instead of making them up30Claude Opus 4.53 of 3 tests
- MathCompetition and research-level math30Claude Opus 4.53 of 3 tests
- ReasoningHard problems that need careful thinking20Claude Opus 4.52 of 3 tests · 1 tie
- Long documentsFinding answers in very long texts10Claude Opus 4.51 of 2 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video05Qwen3.5 27B5 of 5 tests
- Following instructionsDoing exactly what it is asked02Qwen3.5 27B2 of 2 tests
- AgentsCarrying out multi-step tasks on its own22Even2 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.
Qwen3.5 27B costs 91% 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 4.5 pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+25.9points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+20.5points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+14.4points ahead
Where Qwen3.5 27B pulls ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+62.3points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+24points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+17.6points ahead
Every test, side by side
All 57 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 4.5
- SWE-bench MultilingualQwen3.5 27B by 62.3769.3+62.3
- Terminal-Bench HardClaude Opus 4.5 by 14.44732.6+14.4
- LMArena · WebDevClaude Opus 4.5 by 112 rating points14691357+112 rating
- SciCodeClaude Opus 4.5 by 1049.539.5+10
- SWE-bench VerifiedClaude Opus 4.5 by 8.580.972.4+8.5
- LiveCodeBench v6Claude Opus 4.5 by 4.184.880.7+4.1
- SWE-bench Protie5251.2tie
AgentsEven
- BrowseCompQwen3.5 27B by 243761+24
- GDPValClaude Opus 4.5 by 14.347.333+14.3
- OSWorld-VerifiedClaude Opus 4.5 by 10.166.356.2+10.1
- MCP AtlasQwen3.5 27B by 6.162.368.4+6.1
ReasoningClaude Opus 4.5
- Humanity's Last ExamClaude Opus 4.5 by 6.230.123.9+6.2
- CritPtClaude Opus 4.5 by 3.74.60.9+3.7
- GPQA Diamondtie86.685.8tie
FactsClaude Opus 4.5
- AA-Omniscience · AccuracyClaude Opus 4.5 by 25.946.620.7+25.9
- AA-Omniscience · Non-hallucinationClaude Opus 4.5 by 20.53918.5+20.5
- Vectara HHEM hallucination ratelower is betterClaude Opus 4.5 by 1.210.912.1+1.2
Images and chartsQwen3.5 27B
- CharXiv (RQ)Qwen3.5 27B by 12.367.279.5+12.3
- MathVistaQwen3.5 27B by 7.680.287.8+7.6
- OCRBenchQwen3.5 27B by 2.986.589.4+2.9
- MMMUQwen3.5 27B by 1.680.782.3+1.6
- MMMU-ProQwen3.5 27B by 17475+1
MathClaude Opus 4.5
- HMMT Feb. 2026Claude Opus 4.5 by 4.285.381.1+4.2
- IMOAnswerBenchClaude Opus 4.5 by 4.18479.9+4.1
- AIME 2026Claude Opus 4.5 by 3.495.191.7+3.4
Long documentsClaude Opus 4.5
- LongBench v2Claude Opus 4.5 by 3.864.460.6+3.8
- AA-LCRtie77.377.7tie
Following instructionsQwen3.5 27B
- IFBenchQwen3.5 27B by 17.65875.6+17.6
- Multi-ChallengeQwen3.5 27B by 1.85960.8+1.8
Other results28 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 Opus 4.5 by 5814-44+58
- QwenWebBenchClaude Opus 4.5 by 468 rating points15361068+468 rating
- SkillsBench Avg5Claude Opus 4.5 by 18.145.327.2+18.1
- Terminal-Bench 2.0Claude Opus 4.5 by 17.759.341.6+17.7
- NL2RepoClaude Opus 4.5 by 15.943.227.3+15.9
- SimpleVQAClaude Opus 4.5 by 13.769.756+13.7
- Artificial Analysis Coding IndexClaude Opus 4.5 by 12.947.834.9+12.9
- Claw-Eval AvgClaude Opus 4.5 by 12.376.664.3+12.3
- MATH-VisionQwen3.5 27B by 8.977.186+8.9
- ZeroBenchQwen3.5 27B by 7310+7
- AA IntelligenceClaude Opus 4.5 by 6.229.122.9+6.2
- AA Agentic IndexClaude Opus 4.5 by 559.654.6+5
- SuperGPQAClaude Opus 4.5 by 570.665.6+5
- MMMLUClaude Opus 4.5 by 4.990.885.9+4.9
- τ²-Bench Telecom (AA run)Qwen3.5 27B by 4.489.593.9+4.4
- MMVUClaude Opus 4.5 by 477.373.3+4
- HMMT Nov. 2025Claude Opus 4.5 by 3.593.389.8+3.5
- MMLU-ProClaude Opus 4.5 by 3.489.586.1+3.4
- MMLU-ReduxClaude Opus 4.5 by 2.495.693.2+2.4
- VideoMMMUClaude Opus 4.5 by 2.184.482.3+2.1
- vectara_avg_summary_lengthClaude Opus 4.5 by 20.1 rating points114.594.4+20.1 rating
- C-EvalClaude Opus 4.5 by 1.792.290.5+1.7
- OmniDocBench 1.5Qwen3.5 27B by 1.287.788.9+1.2
- vectara_factual_consistencyClaude Opus 4.5 by 1.289.187.9+1.2
- vectara_answer_rateQwen3.5 27B by 1.198.799.8+1.1
- HMMT Feb. 2025tie92.992tie
- Seal-0tie47.747.2tie
- QwenClawBenchtie52.352.2tie
Questions people ask
Which is better, Claude Opus 4.5 or Qwen3.5 27B?
Claude Opus 4.5 wins five of the eight areas we test: coding, reasoning, facts, math and long documents. Qwen3.5 27B wins images and charts and following instructions, and costs 91% less. They are level on agents.
Which is better for coding?
Claude Opus 4.5. It wins 5 of the 7 coding tests both models report; Qwen3.5 27B wins 1, and 1 is a tie.
Which is cheaper?
Claude Opus 4.5 costs $5.00 per million input tokens and $25.00 per million output tokens; Qwen3.5 27B costs $0.30 and $2.40. That makes Qwen3.5 27B about 91% cheaper for the same work.
How do you compare the two?
We use the 57 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 28 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.