Claude Opus 4.8 vs Qwen3.8 Max Preview
Wins 3 of 8 areas
Reasoning · Facts · Math
Wins 4 of 8 areas
Agents · Images and charts · Long documents · Following instructions
Qwen3.8 Max Preview wins more areas, narrowly.Claude Opus 4.8 is better at reasoning.
Both rank among the ten best models we track in agents.
Scores updated · 50 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.
- AgentsCarrying out multi-step tasks on its own05Qwen3.8 Max Preview5 of 5 tests
- Following instructionsDoing exactly what it is asked02Qwen3.8 Max Preview2 of 2 tests
- Images and chartsUnderstanding pictures, charts and video01Qwen3.8 Max Preview1 of 1 test
- Long documentsFinding answers in very long texts01Qwen3.8 Max Preview1 of 1 test
- ReasoningHard problems that need careful thinking30Claude Opus 4.83 of 4 tests · 1 tie
- MathCompetition and research-level math30Claude Opus 4.83 of 3 tests
- FactsGetting facts right instead of making them up21Claude Opus 4.82 of 3 tests
- CodingWriting and fixing software33Even3 each
Images and charts and long documents 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.
Qwen3.8 Max Preview costs 73% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Qwen3.8 Max Preview pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+20.6points ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+17.2points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+14.2points ahead
Where Claude Opus 4.8 pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+17.1points ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+9.8points ahead
- Programming problems, refreshed regularlyLiveBench · Coding+8.9points ahead
Every test, side by side
All 50 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingEven
- LiveBench · Agentic CodingQwen3.8 Max Preview by 14.250.564.7+14.2
- LMArena · WebDevQwen3.8 Max Preview by 116 rating points15561672+116 rating
- LiveBench · CodingClaude Opus 4.8 by 8.981.872.9+8.9
- Terminal-Bench 2.1Qwen3.8 Max Preview by 4.284.688.8+4.2
- SciCodeClaude Opus 4.8 by 2.354.452.1+2.3
- SWE-bench ProClaude Opus 4.8 by 1.569.267.7+1.5
AgentsQwen3.8 Max Preview
- Terminal-Bench 4.0Qwen3.8 Max Preview by 17.221.738.9+17.2
- τ-Bench V3 · BankingQwen3.8 Max Preview by 13.634.247.8+13.6
- GDPValQwen3.8 Max Preview by 10.647.758.3+10.6
- AA ApexAgentsQwen3.8 Max Preview by 339.442.4+3
- OSWorld-VerifiedQwen3.8 Max Preview by 2.783.486.1+2.7
ReasoningClaude Opus 4.8
- Humanity's Last ExamClaude Opus 4.8 by 5.648.743.1+5.6
- CritPtClaude Opus 4.8 by 3.220.917.7+3.2
- LiveBench · ReasoningClaude Opus 4.8 by 189.288.2+1
- GPQA Diamondtie9292.8tie
FactsClaude Opus 4.8
- AA-Omniscience · AccuracyClaude Opus 4.8 by 17.148.831.7+17.1
- AA-Omniscience · Non-hallucinationQwen3.8 Max Preview by 10.560.771.2+10.5
- SimpleQA VerifiedClaude Opus 4.8 by 5.75347.3+5.7
Images and chartsQwen3.8 Max Preview
- LMArena · VisionQwen3.8 Max Preview by 20 rating points12941314+20 rating
MathClaude Opus 4.8
- FrontierMath Tier 4Claude Opus 4.8 by 9.856.146.3+9.8
- FrontierMath Tiers 1-3 (v2)Claude Opus 4.8 by 5.38074.7+5.3
- LiveBench · MathematicsClaude Opus 4.8 by 394.391.3+3
Long documentsQwen3.8 Max Preview
- AA-LCRQwen3.8 Max Preview by 2.677.780.3+2.6
Following instructionsQwen3.8 Max Preview
- IFBenchQwen3.8 Max Preview by 20.662.282.8+20.6
- LiveBench · Instruction FollowingQwen3.8 Max Preview by 2.17274.1+2.1
Other results25 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.
- Agents' Last ExamQwen3.8 Max Preview by 26.725.752.4+26.7
- AA-OmniscienceClaude Opus 4.8 by 16.828.812+16.8
- GDPval-AA v2Qwen3.8 Max Preview by 146 rating points15931739+146 rating
- NL2RepoClaude Opus 4.8 by 13.869.755.9+13.8
- PaperBenchQwen3.8 Max Preview by 12.780.393+12.7
- ToolathlonQwen3.8 Max Preview by 12.659.972.5+12.6
- livebench_data_analysisQwen3.8 Max Preview by 12.46678.4+12.4
- WideSearchQwen3.8 Max Preview by 972.981.9+9
- Toolathlon VerifiedClaude Opus 4.8 by 7.479.972.5+7.4
- Agents' Last Exam (Pass / Score)Qwen3.8 Max Preview by 7.345.152.4+7.3
- AA Agentic IndexQwen3.8 Max Preview by 742.649.6+7
- AndroidBenchQwen3.8 Max Preview by 5.369.875.1+5.3
- JobBenchQwen3.8 Max Preview by 548.453.4+5
- ProgramBench (Almost Solved)Claude Opus 4.8 by 515.510.5+5
- AA IntelligenceQwen3.8 Max Preview by 3.641.845.4+3.6
- ScreenSpot-Pro (No tools)Claude Opus 4.8 by 3.487.984.5+3.4
- Artificial Analysis Coding IndexClaude Opus 4.8 by 2.574.371.8+2.5
- DeepSWE 1.1Claude Opus 4.8 by 2.45956.6+2.4
- MLS Bench Litelower is betterQwen3.8 Max Preview by 1.842.841+1.8
- HLE (with tools)Claude Opus 4.8 by 1.757.956.2+1.7
- DeepSWEClaude Opus 4.8 by 1.45856.6+1.4
- Agents' Last Exam (ALE-CLI)Qwen3.8 Max Preview by 1.325.727+1.3
- HealthBenchtie59.360.2tie
- CyberGymtie78.878.5tie
- livebench_languagetie79.779.7tie
Questions people ask
Which is better, Claude Opus 4.8 or Qwen3.8 Max Preview?
Qwen3.8 Max Preview wins four of the eight areas we test: agents, images and charts, long documents and following instructions. Claude Opus 4.8 wins reasoning, facts and math. They are level on coding.
Which is better for coding?
Neither. They win 3 coding tests each of the 6 both models report.
Which is cheaper?
Claude Opus 4.8 costs $5.00 per million input tokens and $25.00 per million output tokens; Qwen3.8 Max Preview costs $2.00 and $6.00. That makes Qwen3.8 Max Preview about 73% cheaper for the same work.
How do you compare the two?
We use the 50 benchmark tests both models have published scores on. The verdict counts the 25 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 25 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.