o3 vs Qwen3 235B A22B
Wins 6 of 6 areas
Coding · Agents · Reasoning · Facts · Long documents · Following instructions
Wins 0 of 6 areas
—
o3 is the stronger all-rounder.Qwen3 235B A22B is cheaper.
Scores updated · 37 tests both models report · How we compare
Where each one wins
Tests won in each of the six areas where both have results. 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 software30o33 of 3 tests
- ReasoningHard problems that need careful thinking30o33 of 3 tests
- Long documentsFinding answers in very long texts20o32 of 2 tests
- Following instructionsDoing exactly what it is asked20o32 of 2 tests
- FactsGetting facts right instead of making them up21o32 of 3 tests
- AgentsCarrying out multi-step tasks on its own10o31 of 1 test
Agents 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 235B A22B costs 65% 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 o3 pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+34.7points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+32.7points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+31points ahead
Where Qwen3 235B A22B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+10.5points ahead
Every test, side by side
All 37 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Codingo3
- SWE-bench Verifiedo3 by 34.769.134.4+34.7
- Terminal-Bench Hardo3 by 3137.16.1+31
- SciCodeo3 by 1.14139.9+1.1
Reasoningo3
- GPQA Diamondo3 by 12.782.770+12.7
- Humanity's Last Examo3 by 9.120.111+9.1
- CritPto3 by 1.11.10+1.1
Factso3
- AA-Omniscience · Accuracyo3 by 20.138.618.5+20.1
- AA-Omniscience · Non-hallucinationQwen3 235B A22B by 10.511.922.4+10.5
- SimpleQA Verifiedo3 by 949.440.4+9
Long documentso3
- AA-LCRo3 by 74.774.70+74.7
- LongBench v2o3 by 8.758.850.1+8.7
Following instructionso3
- IFBencho3 by 32.771.438.7+32.7
- Multi-Challengeo3 by 19.260.441.2+19.2
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.
- ZebraLogico3 by 58.195.837.7+58.1
- τ²-Bench Telecom (AA run)o3 by 56.780.724+56.7
- Tau2 airlineo3 by 38.364.826.5+38.3
- SimpleQAo3 by 36.249.413.2+36.2
- AA-Omniscienceo3 by 29.2-15.6-44.7+29.2
- OpenAI-MRCR (128k)o3 by 28.856.527.7+28.8
- τ²-Bench (Retail)o3 by 23.280.257+23.2
- Artificial Analysis Coding Indexo3 by 2138.417.4+21
- AA Agentic Indexo3 by 17.736.118.4+17.7
- TAU-bench (airline)o3 by 17.35234.7+17.3
- MMLU-Proo3 by 16.88568.2+16.8
- TAU-bench (retail)o3 by 15.373.958.6+15.3
- HMMT Feb. 2025o3 by 1577.562.5+15
- LiveCodeBencho3 by 1484.770.7+14
- livecodebench_hardo3 by 126654+12
- AA Intelligenceo3 by 10.720.29.5+10.7
- LiveCodeBench (24/8~25/5)o3 by 9.975.865.9+9.9
- MATH-500 (EM)o3 by 6.998.191.2+6.9
- FullStackBencho3 by 6.469.362.9+6.4
- AIME 2024o3 by 5.991.685.7+5.9
- AIME 2025o3 by 4.986.481.5+4.9
- livecodebench_mediumo3 by 189.888.8+1
- livecodebench_easytie99.199.1tie
Questions people ask
Which is better, o3 or Qwen3 235B A22B?
o3 wins all six areas where both have results: coding, agents, reasoning, facts, long documents and following instructions. Qwen3 235B A22B wins none, but costs 65% less.
Which is better for coding?
o3. It wins 3 of the 3 coding tests both models report; Qwen3 235B A22B wins none.
Which is cheaper?
o3 costs $2.00 per million input tokens and $8.00 per million output tokens; Qwen3 235B A22B costs $0.70 and $2.80. That makes Qwen3 235B A22B about 65% cheaper for the same work.
How do you compare the two?
We use the 37 benchmark tests both models have published scores on. The verdict counts the 14 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.