GPT-5.6 Sol vs Qwen3.8 Max Preview
Wins 4 of 8 areas
Reasoning · Facts · Math · Long documents
Wins 4 of 8 areas
Coding · Agents · Images and charts · Following instructions
The two are evenly matched.GPT-5.6 Sol is better at reasoning and math; Qwen3.8 Max Preview at agents and following instructions.
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.
- ReasoningHard problems that need careful thinking40GPT-5.6 Sol4 of 4 tests
- MathCompetition and research-level math30GPT-5.6 Sol3 of 3 tests
- FactsGetting facts right instead of making them up21GPT-5.6 Sol2 of 3 tests
- Long documentsFinding answers in very long texts10GPT-5.6 Sol1 of 1 test
- AgentsCarrying out multi-step tasks on its own24Qwen3.8 Max Preview4 of 6 tests
- Following instructionsDoing exactly what it is asked02Qwen3.8 Max Preview2 of 2 tests
- CodingWriting and fixing software23Qwen3.8 Max Preview3 of 6 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video01Qwen3.8 Max Preview1 of 2 tests · 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.
Qwen3.8 Max Preview costs 67% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where GPT-5.6 Sol pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+36.7points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+27.7points ahead
- Short factual questions, answered correctlySimpleQA Verified+22.4points ahead
Where Qwen3.8 Max Preview pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+63.4points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+10.1points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+8.5points 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.
CodingQwen3.8 Max Preview
- LiveBench · CodingGPT-5.6 Sol by 1183.972.9+11
- LiveBench · Agentic CodingQwen3.8 Max Preview by 8.556.264.7+8.5
- LMArena · WebDevQwen3.8 Max Preview by 56 rating points16161672+56 rating
- SciCodeGPT-5.6 Sol by 557.152.1+5
- SWE-bench ProQwen3.8 Max Preview by 3.164.667.7+3.1
- Terminal-Bench 2.1tie8888.8tie
AgentsQwen3.8 Max Preview
- AA IT-Bench SREGPT-5.6 Sol by 15.956.240.3+15.9
- τ-Bench V3 · BankingQwen3.8 Max Preview by 3.544.347.8+3.5
- OSWorld-VerifiedQwen3.8 Max Preview by 3.18386.1+3.1
- GDPValQwen3.8 Max Preview by 2.955.458.3+2.9
- AA ApexAgentsQwen3.8 Max Preview by 2.539.942.4+2.5
- Terminal-Bench 4.0GPT-5.6 Sol by 139.938.9+1
ReasoningGPT-5.6 Sol
- CritPtGPT-5.6 Sol by 14.632.317.7+14.6
- Humanity's Last ExamGPT-5.6 Sol by 6.449.543.1+6.4
- LiveBench · ReasoningGPT-5.6 Sol by 3.591.788.2+3.5
- GPQA DiamondGPT-5.6 Sol by 1.394.192.8+1.3
FactsGPT-5.6 Sol
- AA-Omniscience · Non-hallucinationQwen3.8 Max Preview by 63.47.871.2+63.4
- AA-Omniscience · AccuracyGPT-5.6 Sol by 27.759.431.7+27.7
- SimpleQA VerifiedGPT-5.6 Sol by 22.469.747.3+22.4
Images and chartsQwen3.8 Max Preview
- LMArena · VisionQwen3.8 Max Preview by 33 rating points12811314+33 rating
- MMMU-Protie83.482.8tie
MathGPT-5.6 Sol
- FrontierMath Tier 4GPT-5.6 Sol by 36.78346.3+36.7
- FrontierMath Tiers 1-3 (v2)GPT-5.6 Sol by 14.489.174.7+14.4
- LiveBench · MathematicsGPT-5.6 Sol by 4.996.291.3+4.9
Following instructionsQwen3.8 Max Preview
- IFBenchQwen3.8 Max Preview by 10.172.782.8+10.1
- LiveBench · Instruction FollowingQwen3.8 Max Preview by 2.271.874.1+2.2
Other results23 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.
- DeepSWE 1.1GPT-5.6 Sol by 16.47356.6+16.4
- DeepSWEGPT-5.6 Sol by 16.172.756.6+16.1
- ToolathlonQwen3.8 Max Preview by 14.55872.5+14.5
- ProgramBench (Almost Solved)GPT-5.6 Sol by 12.52310.5+12.5
- AA-OmniscienceGPT-5.6 Sol by 102212+10
- HLE (with tools)GPT-5.6 Sol by 8.364.556.2+8.3
- JobBenchQwen3.8 Max Preview by 845.453.4+8
- livebench_languageGPT-5.6 Sol by 887.779.7+8
- CyberGymGPT-5.6 Sol by 684.578.5+6
- Artificial Analysis Coding IndexGPT-5.6 Sol by 5.677.471.8+5.6
- MLS Bench Litelower is betterQwen3.8 Max Preview by 5.246.241+5.2
- HealthBenchQwen3.8 Max Preview by 3.25760.2+3.2
- GDPval-AA v2Qwen3.8 Max Preview by 28 rating points17111739+28 rating
- PaperBenchQwen3.8 Max Preview by 2.590.593+2.5
- Toolathlon VerifiedGPT-5.6 Sol by 2.474.972.5+2.4
- AA IntelligenceGPT-5.6 Sol by 1.64745.4+1.6
- Agents' Last Exam (ALE-CLI)GPT-5.6 Sol by 1.628.627+1.6
- livebench_data_analysisGPT-5.6 Sol by 1.479.878.4+1.4
- Agents' Last Exam (Pass / Score)GPT-5.6 Sol by 1.253.652.4+1.2
- AndroidBenchQwen3.8 Max Preview by 1.17475.1+1.1
- AA Agentic Indextie50.549.6tie
- Agents' Last Examtie52.752.4tie
- LVBenchtie82.181.8tie
Questions people ask
Which is better, GPT-5.6 Sol or Qwen3.8 Max Preview?
GPT-5.6 Sol and Qwen3.8 Max Preview each win four of the eight areas we test. GPT-5.6 Sol wins reasoning, facts, math and long documents; Qwen3.8 Max Preview wins coding, agents, images and charts and following instructions. Qwen3.8 Max Preview costs 67% less.
Which is better for coding?
Qwen3.8 Max Preview. It wins 3 of the 6 coding tests both models report; GPT-5.6 Sol wins 2, and 1 is a tie.
Which is cheaper?
GPT-5.6 Sol costs $4.00 per million input tokens and $20.00 per million output tokens; Qwen3.8 Max Preview costs $2.00 and $6.00. That makes Qwen3.8 Max Preview about 67% 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 27 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 23 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.