GPT-5.4 vs Qwen3.8 27B
Wins 3 of 8 areas
Reasoning · Images and charts · Math
Wins 3 of 8 areas
Coding · Agents · Following instructions
The two are evenly matched.GPT-5.4 is better at reasoning and images and charts; Qwen3.8 27B at coding and agents.
Scores updated · 39 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 software25Qwen3.8 27B5 of 7 tests
- AgentsCarrying out multi-step tasks on its own14Qwen3.8 27B4 of 5 tests
- Following instructionsDoing exactly what it is asked02Qwen3.8 27B2 of 2 tests
- ReasoningHard problems that need careful thinking50GPT-5.45 of 5 tests
- Images and chartsUnderstanding pictures, charts and video21GPT-5.42 of 3 tests
- MathCompetition and research-level math10GPT-5.41 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
- Long documentsFinding answers in very long texts00Even0 each · 1 tie
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 80% 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 27B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+61.4points ahead
- Multi-step tasks using many real software toolsMCP Atlas+10.1points ahead
- Completes tasks by operating a computer desktopOSWorld-Verified+9.3points ahead
Where GPT-5.4 pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+35.3points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+30.9points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+21.1points ahead
Every test, side by side
All 39 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.8 27B
- LMArena · WebDevQwen3.8 27B by 128 rating points14621590+128 rating
- SciCodeGPT-5.4 by 1056.646.6+10
- LiveBench · Agentic CodingQwen3.8 27B by 7.653.861.4+7.6
- SWE-bench ProQwen3.8 27B by 457.761.7+4
- SWE-bench MultilingualQwen3.8 27B by 2.171.773.8+2.1
- LiveBench · CodingGPT-5.4 by 1.877.575.7+1.8
- Terminal-Bench 2.1Qwen3.8 27B by 1.578.379.8+1.5
AgentsQwen3.8 27B
- BrowseCompGPT-5.4 by 21.182.761.6+21.1
- MCP AtlasQwen3.8 27B by 10.167.277.3+10.1
- OSWorld-VerifiedQwen3.8 27B by 9.37584.3+9.3
- GDPValQwen3.8 27B by 8.737.346+8.7
- τ-Bench V3 · BankingQwen3.8 27B by 8.439.648+8.4
ReasoningGPT-5.4
- ARC-AGI-2GPT-5.4 by 30.973.342.4+30.9
- CritPtGPT-5.4 by 1823.45.4+18
- Humanity's Last ExamGPT-5.4 by 9.843.733.9+9.8
- LiveBench · ReasoningGPT-5.4 by 8.188.180+8.1
- GPQA DiamondGPT-5.4 by 1.59290.5+1.5
FactsEven
- AA-Omniscience · Non-hallucinationQwen3.8 27B by 61.48.369.7+61.4
- AA-Omniscience · AccuracyGPT-5.4 by 35.350.915.6+35.3
Images and chartsGPT-5.4
- CharXiv (RQ)Qwen3.8 27B by 7.482.890.2+7.4
- LMArena · VisionGPT-5.4 by 30 rating points13011271+30 rating
- MMMU-ProGPT-5.4 by 2.178.476.3+2.1
Following instructionsQwen3.8 27B
- IFBenchQwen3.8 27B by 5.673.979.5+5.6
- LiveBench · Instruction FollowingQwen3.8 27B by 2.570.272.7+2.5
Other results13 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.
- BabyVisionQwen3.8 27B by 35.949.785.6+35.9
- AA-OmniscienceGPT-5.4 by 15.85.8-10+15.8
- DeepSWE 1.1GPT-5.4 by 9.85242.2+9.8
- livebench_languageGPT-5.4 by 8.282.674.3+8.2
- ARC-AGI-1GPT-5.4 by 6.293.787.5+6.2
- AA IntelligenceGPT-5.4 by 5.33933.7+5.3
- Artificial Analysis Coding IndexGPT-5.4 by 371.168.1+3
- livebench_data_analysisGPT-5.4 by 2.779.376.6+2.7
- MATH-VisionQwen3.8 27B by 2.69294.6+2.6
- AA Agentic IndexQwen3.8 27B by 2.344.246.5+2.3
- OmniDocBench 1.5Qwen3.8 27B by 289.191.1+2
- DeepSearchQA (F1)Qwen3.8 27B by 1.178.679.7+1.1
- NL2RepoQwen3.8 27B by 141.342.3+1
Questions people ask
Which is better, GPT-5.4 or Qwen3.8 27B?
GPT-5.4 and Qwen3.8 27B each win three of the eight areas we test. GPT-5.4 wins reasoning, images and charts and math; Qwen3.8 27B wins coding, agents and following instructions. Qwen3.8 27B costs 80% less. They are level on facts and long documents.
Which is better for coding?
Qwen3.8 27B. It wins 5 of the 7 coding tests both models report; GPT-5.4 wins 2.
Which is cheaper?
GPT-5.4 costs $2.50 per million input tokens and $15.00 per million output tokens; Qwen3.8 27B costs $0.50 and $3.00. That makes Qwen3.8 27B about 80% cheaper for the same work.
How do you compare the two?
We use the 39 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 13 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.