GPT-5.4 vs Qwen3.5 27B
Wins 7 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Long documents
Wins 1 of 8 areas
Following instructions
GPT-5.4 is the stronger all-rounder.Qwen3.5 27B is cheaper and better at following instructions.
Scores updated · 42 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 software50GPT-5.45 of 5 tests
- ReasoningHard problems that need careful thinking30GPT-5.43 of 3 tests
- Images and chartsUnderstanding pictures, charts and video30GPT-5.43 of 3 tests
- MathCompetition and research-level math30GPT-5.43 of 3 tests
- AgentsCarrying out multi-step tasks on its own32GPT-5.43 of 5 tests
- FactsGetting facts right instead of making them up21GPT-5.42 of 3 tests
- Long documentsFinding answers in very long texts10GPT-5.41 of 1 test
- Following instructionsDoing exactly what it is asked01Qwen3.5 27B1 of 1 test
Long documents and following instructions 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.
Qwen3.5 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-5.4 pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+30.2points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+25points ahead
- Research-level physics problemsCritPt+22.5points ahead
Where Qwen3.5 27B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+10.2points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+1.7points ahead
- Multi-step tasks using many real software toolsMCP Atlas+1.2points ahead
Every test, side by side
All 42 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.4
- Terminal-Bench HardGPT-5.4 by 2557.632.6+25
- SciCodeGPT-5.4 by 17.156.639.5+17.1
- LMArena · WebDevGPT-5.4 by 108 rating points14651357+108 rating
- SWE-bench ProGPT-5.4 by 6.557.751.2+6.5
- SWE-bench MultilingualGPT-5.4 by 2.471.769.3+2.4
AgentsGPT-5.4
- BrowseCompGPT-5.4 by 21.782.761+21.7
- OSWorld-VerifiedGPT-5.4 by 18.87556.2+18.8
- GDPValGPT-5.4 by 4.437.433+4.4
- MCP AtlasQwen3.5 27B by 1.267.268.4+1.2
- AA IT-Bench SREQwen3.5 27B by 134.535.5+1
ReasoningGPT-5.4
- CritPtGPT-5.4 by 22.523.40.9+22.5
- Humanity's Last ExamGPT-5.4 by 19.843.723.9+19.8
- GPQA DiamondGPT-5.4 by 6.29285.8+6.2
FactsGPT-5.4
- AA-Omniscience · AccuracyGPT-5.4 by 30.250.920.7+30.2
- AA-Omniscience · Non-hallucinationQwen3.5 27B by 10.28.318.5+10.2
- Vectara HHEM hallucination ratelower is betterGPT-5.4 by 5.1712.1+5.1
Images and chartsGPT-5.4
- LMArena · VisionGPT-5.4 by 62 rating points13021240+62 rating
- MMMU-ProGPT-5.4 by 3.478.475+3.4
- CharXiv (RQ)GPT-5.4 by 3.382.879.5+3.3
MathGPT-5.4
- HMMT Feb. 2026GPT-5.4 by 16.697.781.1+16.6
- IMOAnswerBenchGPT-5.4 by 11.591.479.9+11.5
- AIME 2026GPT-5.4 by 7.599.291.7+7.5
Following instructionsQwen3.5 27B
- IFBenchQwen3.5 27B by 1.773.975.6+1.7
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-5.4 by 49.85.8-44+49.8
- Artificial Analysis Coding IndexGPT-5.4 by 36.271.134.9+36.2
- Terminal-Bench 2.0GPT-5.4 by 33.575.141.6+33.5
- MCPMarkGPT-5.4 by 26.262.536.3+26.2
- Tool-DecathlonGPT-5.4 by 23.154.631.5+23.1
- AA IntelligenceGPT-5.4 by 16.13922.9+16.1
- NL2RepoGPT-5.4 by 1441.327.3+14
- vectara_avg_summary_lengthQwen3.5 27B by 12.781.794.4+12.7
- AA Agentic IndexQwen3.5 27B by 10.444.254.6+10.4
- τ²-Bench Telecom (AA run)Qwen3.5 27B by 6.887.193.9+6.8
- HMMT Nov. 2025GPT-5.4 by 695.889.8+6
- MATH-VisionGPT-5.4 by 69286+6
- BabyVisionGPT-5.4 by 5.149.744.6+5.1
- vectara_factual_consistencyGPT-5.4 by 5.19387.9+5.1
- τ³-BenchGPT-5.4 by 4.572.968.4+4.5
- MMLU-ProGPT-5.4 by 1.487.586.1+1.4
- OmniDocBench 1.5tie89.188.9tie
- vectara_answer_ratetie99.999.8tie
Questions people ask
Which is better, GPT-5.4 or Qwen3.5 27B?
GPT-5.4 wins seven of the eight areas we test: coding, agents, reasoning, facts, images and charts, math and long documents. Qwen3.5 27B wins following instructions, and costs 85% less.
Which is better for coding?
GPT-5.4. It wins 5 of the 5 coding tests both models report; Qwen3.5 27B wins none.
Which is cheaper?
GPT-5.4 costs $2.50 per million input tokens and $15.00 per million output tokens; Qwen3.5 27B costs $0.30 and $2.40. That makes Qwen3.5 27B about 85% cheaper for the same work.
How do you compare the two?
We use the 42 benchmark tests both models have published scores on. The verdict counts the 24 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.