Claude Opus 4.6 vs Qwen3.5 35B A3B
Wins 6 of 8 areas
Coding · Agents · Reasoning · Facts · Math · Long documents
Wins 1 of 8 areas
Following instructions
Claude Opus 4.6 is the stronger all-rounder.Qwen3.5 35B A3B is cheaper and better at following instructions.
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 software70Claude Opus 4.67 of 7 tests
- AgentsCarrying out multi-step tasks on its own30Claude Opus 4.63 of 4 tests · 1 tie
- ReasoningHard problems that need careful thinking30Claude Opus 4.63 of 3 tests
- FactsGetting facts right instead of making them up21Claude Opus 4.62 of 3 tests
- MathCompetition and research-level math21Claude Opus 4.62 of 3 tests
- Long documentsFinding answers in very long texts10Claude Opus 4.61 of 1 test
- Following instructionsDoing exactly what it is asked02Qwen3.5 35B A3B2 of 2 tests
- Images and chartsUnderstanding pictures, charts and video11Even1 each · 1 tie
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.
Qwen3.5 35B A3B costs 93% 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
- Real work tasks from 44 professionsGDPVal+50.3points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+26.9points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+23points ahead
Where Qwen3.5 35B A3B pulls ahead
- Keeps track of context across a multi-turn chatMulti-Challenge+22.8points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+19.4points ahead
- Reads and reasons about text in imagesOCRBenchv2+16.9points 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 4.6
- LMArena · WebDevClaude Opus 4.6 by 296 rating points15461250+296 rating
- Terminal-Bench HardClaude Opus 4.6 by 19.746.226.5+19.7
- SWE-bench MultilingualClaude Opus 4.6 by 17.577.860.3+17.5
- LiveCodeBench v6Claude Opus 4.6 by 14.288.874.6+14.2
- SciCodeClaude Opus 4.6 by 14.251.937.7+14.2
- SWE-bench VerifiedClaude Opus 4.6 by 11.680.869.2+11.6
- SWE-bench ProClaude Opus 4.6 by 8.853.444.6+8.8
AgentsClaude Opus 4.6
- GDPValClaude Opus 4.6 by 50.355.95.6+50.3
- BrowseCompClaude Opus 4.6 by 238461+23
- OSWorld-VerifiedClaude Opus 4.6 by 18.272.754.5+18.2
- MCP Atlastie62.762.4tie
ReasoningClaude Opus 4.6
- Humanity's Last ExamClaude Opus 4.6 by 18.939.921+18.9
- CritPtClaude Opus 4.6 by 11.712.60.9+11.7
- GPQA DiamondClaude Opus 4.6 by 5.189.684.5+5.1
FactsClaude Opus 4.6
- AA-Omniscience · AccuracyClaude Opus 4.6 by 26.94720.1+26.9
- AA-Omniscience · Non-hallucinationClaude Opus 4.6 by 22.637.214.6+22.6
- Vectara HHEM hallucination ratelower is betterQwen3.5 35B A3B by 1.712.210.5+1.7
Images and chartsEven
- OCRBenchv2Qwen3.5 35B A3B by 16.948.465.3+16.9
- MMMU-ProClaude Opus 4.6 by 2.775.472.7+2.7
- CharXiv (RQ)tie77.477.5tie
MathClaude Opus 4.6
- HMMT Feb. 2026Claude Opus 4.6 by 14.496.281.8+14.4
- AIME 2026Claude Opus 4.6 by 3.496.793.3+3.4
- IMOAnswerBenchQwen3.5 35B A3B by 1.575.376.8+1.5
Following instructionsQwen3.5 35B A3B
- Multi-ChallengeQwen3.5 35B A3B by 22.837.160+22.8
- IFBenchQwen3.5 35B A3B by 19.453.172.5+19.4
Other results31 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 4.6 by 61.813.7-48.1+61.8
- AA Agentic IndexClaude Opus 4.6 by 55.867.611.8+55.8
- MCPMarkClaude Opus 4.6 by 29.756.727+29.7
- NL2RepoClaude Opus 4.6 by 29.349.820.5+29.3
- Terminal-Bench 2.0Claude Opus 4.6 by 24.965.440.5+24.9
- ERQAQwen3.5 35B A3B by 2440.864.8+24
- BabyVisionQwen3.5 35B A3B by 23.614.838.4+23.6
- Tool-DecathlonClaude Opus 4.6 by 18.547.228.7+18.5
- MATH-VisionQwen3.5 35B A3B by 18.465.583.9+18.4
- AgentWorldBench - SWEClaude Opus 4.6 by 1764.547.6+17
- AgentWorldBench - OSClaude Opus 4.6 by 13.970.256.3+13.9
- AA IntelligenceClaude Opus 4.6 by 12.631.919.3+12.6
- AgentWorldBench - MCPClaude Opus 4.6 by 1269.957.9+12
- AgentWorldBench - TerminalClaude Opus 4.6 by 11.457.546.1+11.4
- Artificial Analysis Coding IndexClaude Opus 4.6 by 11.148.137+11.1
- ScreenSpot-Pro (No tools)Qwen3.5 35B A3B by 10.957.768.6+10.9
- RealWorldQAQwen3.5 35B A3B by 10.273.984.1+10.2
- AgentWorldBench - OverallClaude Opus 4.6 by 10.157.847.7+10.1
- AgentWorldBench - AndroidClaude Opus 4.6 by 8.561.753.2+8.5
- LVBenchQwen3.5 35B A3B by 8.46371.4+8.4
- HMMT Nov. 2025Claude Opus 4.6 by 7.196.389.2+7.1
- MMMLUClaude Opus 4.6 by 5.991.185.2+5.9
- AgentWorldBench - WebClaude Opus 4.6 by 4.351.447.1+4.3
- vectara_avg_summary_lengthClaude Opus 4.6 by 42.7 rating points137.694.9+42.7 rating
- MMLU-ProClaude Opus 4.6 by 3.889.185.3+3.8
- τ³-BenchClaude Opus 4.6 by 3.572.468.9+3.5
- AgentWorldBench - SearchClaude Opus 4.6 by 3.329.326+3.3
- τ²-Bench Telecom (AA run)Claude Opus 4.6 by 2.992.189.2+2.9
- OCRBench v2 (Chinese)Qwen3.5 35B A3B by 1.959.861.7+1.9
- vectara_factual_consistencyQwen3.5 35B A3B by 1.787.889.5+1.7
- vectara_answer_ratetie99.899.8tie
Questions people ask
Which is better, Claude Opus 4.6 or Qwen3.5 35B A3B?
Claude Opus 4.6 wins six of the eight areas we test: coding, agents, reasoning, facts, math and long documents. Qwen3.5 35B A3B wins following instructions, and costs 93% less. They are level on images and charts.
Which is better for coding?
Claude Opus 4.6. It wins 7 of the 7 coding tests both models report; Qwen3.5 35B A3B wins none.
Which is cheaper?
Claude Opus 4.6 costs $5.00 per million input tokens and $25.00 per million output tokens; Qwen3.5 35B A3B costs $0.25 and $2.00. That makes Qwen3.5 35B A3B about 93% 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 26 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 31 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.