GPT-5.4 vs Qwen3.5 35B A3B
Wins 8 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Long documents · Following instructions
Wins 0 of 8 areas
—
GPT-5.4 is the stronger all-rounder.Qwen3.5 35B A3B is cheaper.
Scores updated · 51 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 software60GPT-5.46 of 6 tests
- AgentsCarrying out multi-step tasks on its own60GPT-5.46 of 6 tests
- ReasoningHard problems that need careful thinking30GPT-5.43 of 3 tests
- MathCompetition and research-level math30GPT-5.43 of 3 tests
- Images and chartsUnderstanding pictures, charts and video20GPT-5.42 of 2 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 asked10GPT-5.41 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 35B A3B costs 87% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where GPT-5.4 pulls ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+37.5points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+34.7points ahead
- Real work tasks from 44 professionsGDPVal+31.8points ahead
Where Qwen3.5 35B A3B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+6.3points ahead
Every test, side by side
All 51 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.4
- Terminal-Bench 2.1GPT-5.4 by 37.578.340.8+37.5
- Terminal-Bench HardGPT-5.4 by 31.157.626.5+31.1
- LMArena · WebDevGPT-5.4 by 215 rating points14651250+215 rating
- SciCodeGPT-5.4 by 18.956.637.7+18.9
- SWE-bench ProGPT-5.4 by 13.157.744.6+13.1
- SWE-bench MultilingualGPT-5.4 by 11.471.760.3+11.4
AgentsGPT-5.4
- τ-Bench V3 · BankingGPT-5.4 by 34.739.64.9+34.7
- GDPValGPT-5.4 by 31.837.45.6+31.8
- BrowseCompGPT-5.4 by 21.782.761+21.7
- OSWorld-VerifiedGPT-5.4 by 20.57554.5+20.5
- AA IT-Bench SREGPT-5.4 by 1334.521.5+13
- MCP AtlasGPT-5.4 by 4.867.262.4+4.8
ReasoningGPT-5.4
- Humanity's Last ExamGPT-5.4 by 22.743.721+22.7
- CritPtGPT-5.4 by 22.523.40.9+22.5
- GPQA DiamondGPT-5.4 by 7.59284.5+7.5
FactsGPT-5.4
- AA-Omniscience · AccuracyGPT-5.4 by 30.850.920.1+30.8
- AA-Omniscience · Non-hallucinationQwen3.5 35B A3B by 6.38.314.6+6.3
- Vectara HHEM hallucination ratelower is betterGPT-5.4 by 3.5710.5+3.5
Images and chartsGPT-5.4
- MMMU-ProGPT-5.4 by 5.778.472.7+5.7
- CharXiv (RQ)GPT-5.4 by 5.382.877.5+5.3
MathGPT-5.4
- HMMT Feb. 2026GPT-5.4 by 15.997.781.8+15.9
- IMOAnswerBenchGPT-5.4 by 14.691.476.8+14.6
- AIME 2026GPT-5.4 by 5.999.293.3+5.9
Other results26 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 53.95.8-48.1+53.9
- MCPMarkGPT-5.4 by 35.562.527+35.5
- Terminal-Bench 2.0GPT-5.4 by 34.675.140.5+34.6
- Artificial Analysis Coding IndexGPT-5.4 by 34.171.137+34.1
- AA Agentic IndexGPT-5.4 by 32.444.211.8+32.4
- Tool-DecathlonGPT-5.4 by 25.954.628.7+25.9
- NL2RepoGPT-5.4 by 20.841.320.5+20.8
- AA IntelligenceGPT-5.4 by 19.73919.3+19.7
- AgentWorldBench - SWEGPT-5.4 by 18.766.347.6+18.7
- vectara_avg_summary_lengthQwen3.5 35B A3B by 13.281.794.9+13.2
- AgentWorldBench - OSGPT-5.4 by 12.368.656.3+12.3
- AgentWorldBench - MCPGPT-5.4 by 12.270.157.9+12.2
- AgentWorldBench - SearchGPT-5.4 by 11.337.326+11.3
- BabyVisionGPT-5.4 by 11.349.738.4+11.3
- AgentWorldBench - OverallGPT-5.4 by 10.658.347.7+10.6
- MATH-VisionGPT-5.4 by 8.19283.9+8.1
- AgentWorldBench - TerminalGPT-5.4 by 7.653.746.1+7.6
- AgentWorldBench - AndroidGPT-5.4 by 6.86053.2+6.8
- HMMT Nov. 2025GPT-5.4 by 6.695.889.2+6.6
- AgentWorldBench - WebGPT-5.4 by 4.751.847.1+4.7
- τ³-BenchGPT-5.4 by 472.968.9+4
- vectara_factual_consistencyGPT-5.4 by 3.59389.5+3.5
- MMLU-ProGPT-5.4 by 2.287.585.3+2.2
- τ²-Bench Telecom (AA run)Qwen3.5 35B A3B by 2.187.189.2+2.1
- OmniDocBench 1.5tie89.189.3tie
- vectara_answer_ratetie99.999.8tie
Questions people ask
Which is better, GPT-5.4 or Qwen3.5 35B A3B?
GPT-5.4 wins all eight areas we test: coding, agents, reasoning, facts, images and charts, math, long documents and following instructions. Qwen3.5 35B A3B wins none, but costs 87% less.
Which is better for coding?
GPT-5.4. It wins 6 of the 6 coding tests both models report; Qwen3.5 35B A3B wins none.
Which is cheaper?
GPT-5.4 costs $2.50 per million input tokens and $15.00 per million output tokens; Qwen3.5 35B A3B costs $0.25 and $2.00. That makes Qwen3.5 35B A3B about 87% cheaper for the same work.
How do you compare the two?
We use the 51 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 26 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.