GPT-5.6 Sol vs Qwen3.8 27B
Wins 6 of 8 areas
Coding · Agents · Reasoning · Images and charts · Math · Long documents
Wins 1 of 8 areas
Following instructions
GPT-5.6 Sol is the stronger all-rounder.Qwen3.8 27B is cheaper and better at following instructions.
Scores updated · 45 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 thinking60GPT-5.6 Sol6 of 6 tests
- CodingWriting and fixing software51GPT-5.6 Sol5 of 6 tests
- AgentsCarrying out multi-step tasks on its own42GPT-5.6 Sol4 of 6 tests
- Images and chartsUnderstanding pictures, charts and video21GPT-5.6 Sol2 of 3 tests
- MathCompetition and research-level math10GPT-5.6 Sol1 of 1 test
- Long documentsFinding answers in very long texts10GPT-5.6 Sol1 of 1 test
- Following instructionsDoing exactly what it is asked01Qwen3.8 27B1 of 2 tests · 1 tie
- FactsGetting facts right instead of making them up11Even1 each
Math 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 27B costs 85% 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
- Abstract visual puzzles that people can solveARC-AGI-2+50.1points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+43.8points ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+34.3points ahead
Where Qwen3.8 27B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+61.9points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+6.8points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+5.2points ahead
Every test, side by side
All 45 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.6 Sol
- SciCodeGPT-5.6 Sol by 10.557.146.6+10.5
- LiveBench · CodingGPT-5.6 Sol by 8.283.975.7+8.2
- Terminal-Bench 2.1GPT-5.6 Sol by 8.28879.8+8.2
- LiveBench · Agentic CodingQwen3.8 27B by 5.256.261.4+5.2
- SWE-bench ProGPT-5.6 Sol by 2.964.661.7+2.9
- LMArena · WebDevGPT-5.6 Sol by 25 rating points16151590+25 rating
AgentsGPT-5.6 Sol
- Terminal-Bench 4.0GPT-5.6 Sol by 34.339.95.6+34.3
- BrowseCompGPT-5.6 Sol by 28.890.461.6+28.8
- GDPValGPT-5.6 Sol by 9.455.446+9.4
- MCP AtlasGPT-5.6 Sol by 6.383.677.3+6.3
- τ-Bench V3 · BankingQwen3.8 27B by 3.744.348+3.7
- OSWorld-VerifiedQwen3.8 27B by 1.38384.3+1.3
ReasoningGPT-5.6 Sol
- ARC-AGI-2GPT-5.6 Sol by 50.192.542.4+50.1
- CritPtGPT-5.6 Sol by 26.932.35.4+26.9
- Humanity's Last ExamGPT-5.6 Sol by 15.649.533.9+15.6
- LiveBench · ReasoningGPT-5.6 Sol by 11.791.780+11.7
- SimpleBenchGPT-5.6 Sol by 4.664.860.2+4.6
- GPQA DiamondGPT-5.6 Sol by 3.694.190.5+3.6
FactsEven
- AA-Omniscience · Non-hallucinationQwen3.8 27B by 61.97.869.7+61.9
- AA-Omniscience · AccuracyGPT-5.6 Sol by 43.859.415.6+43.8
Images and chartsGPT-5.6 Sol
- MMMU-ProGPT-5.6 Sol by 7.183.476.3+7.1
- CharXiv (RQ)Qwen3.8 27B by 4.485.890.2+4.4
- LMArena · VisionGPT-5.6 Sol by 11 rating points12821271+11 rating
Following instructionsQwen3.8 27B
- IFBenchQwen3.8 27B by 6.872.779.5+6.8
- LiveBench · Instruction Followingtie71.872.7tie
Other results18 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-OmniscienceGPT-5.6 Sol by 3222-10+32
- Terminal-Bench 4.0GPT-5.6 Sol by 31.737.35.6+31.7
- DeepSWE 1.1GPT-5.6 Sol by 30.572.742.2+30.5
- Terminal-Bench-Science 0.1GPT-5.6 Sol by 22.422.40+22.4
- GDPval-AA 2.1GPT-5.6 Sol by 161 rating points15881427+161 rating
- OSWorld 2.0 (partial)GPT-5.6 Sol by 14.662.648+14.6
- GDP (Surge AI)GPT-5.6 Sol by 14.330.716.4+14.3
- AA IntelligenceGPT-5.6 Sol by 13.34733.7+13.3
- livebench_languageGPT-5.6 Sol by 13.387.774.3+13.3
- JobBenchGPT-5.6 Sol by 1245.433.4+12
- Agents' Last ExamGPT-5.6 Sol by 9.852.742.9+9.8
- LVBenchGPT-5.6 Sol by 9.782.172.4+9.7
- Artificial Analysis Coding IndexGPT-5.6 Sol by 9.377.468.1+9.3
- ARC-AGI-1GPT-5.6 Sol by 996.587.5+9
- Toolathlon VerifiedGPT-5.6 Sol by 7.874.967.1+7.8
- AA Agentic IndexGPT-5.6 Sol by 450.546.5+4
- BabyVisionGPT-5.6 Sol by 3.388.985.6+3.3
- livebench_data_analysisGPT-5.6 Sol by 3.279.876.6+3.2
Questions people ask
Which is better, GPT-5.6 Sol or Qwen3.8 27B?
GPT-5.6 Sol wins six of the eight areas we test: coding, agents, reasoning, images and charts, math and long documents. Qwen3.8 27B wins following instructions, and costs 85% less. They are level on facts.
Which is better for coding?
GPT-5.6 Sol. It wins 5 of the 6 coding tests both models report; Qwen3.8 27B wins 1.
Which is cheaper?
GPT-5.6 Sol costs $4.00 per million input tokens and $20.00 per million output tokens; Qwen3.8 27B costs $0.50 and $3.00. That makes Qwen3.8 27B about 85% cheaper for the same work.
How do you compare the two?
We use the 45 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 18 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.