DeepSeek-R1 vs Qwen3 235B A22B Instruct 2507
Wins 1 of 6 areas
Long documents
Wins 3 of 6 areas
Agents · Reasoning · Following instructions
Qwen3 235B A22B Instruct 2507 wins more areas, narrowly.DeepSeek-R1 is better at long documents.
Scores updated · 29 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.
- ReasoningHard problems that need careful thinking02Qwen3 235B A22B Instruct 25072 of 3 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own01Qwen3 235B A22B Instruct 25071 of 1 test
- Following instructionsDoing exactly what it is asked01Qwen3 235B A22B Instruct 25071 of 1 test
- Long documentsFinding answers in very long texts10DeepSeek-R11 of 1 test
- CodingWriting and fixing software11Even1 each
- FactsGetting facts right instead of making them up11Even1 each
Agents, 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 235B A22B Instruct 2507 costs 81% 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 235B A22B Instruct 2507 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+13.1points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+9.1points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+7.1points ahead
Where DeepSeek-R1 pulls ahead
- Reasons across sets of long documentsAA-LCR+23.8points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+11.8points ahead
- Code for real scientific research problemsSciCode+2.3points ahead
Every test, side by side
All 29 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingEven
- Terminal-Bench HardQwen3 235B A22B Instruct 2507 by 9.16.115.2+9.1
- SciCodeDeepSeek-R1 by 2.338.336+2.3
AgentsQwen3 235B A22B Instruct 2507
- GDPValQwen3 235B A22B Instruct 2507 by 14.1014.1+14.1
ReasoningQwen3 235B A22B Instruct 2507
- GPQA DiamondQwen3 235B A22B Instruct 2507 by 4.570.875.3+4.5
- Humanity's Last ExamQwen3 235B A22B Instruct 2507 by 2.68.511.1+2.6
- CritPttie0.60tie
FactsEven
- AA-Omniscience · Non-hallucinationQwen3 235B A22B Instruct 2507 by 13.19.722.8+13.1
- AA-Omniscience · AccuracyDeepSeek-R1 by 11.830.518.7+11.8
Following instructionsQwen3 235B A22B Instruct 2507
- IFBenchQwen3 235B A22B Instruct 2507 by 7.13946.1+7.1
Other results19 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.
- HarmBenchQwen3 235B A22B Instruct 2507 by 31.747.979.6+31.7
- AIR-Bench 2024Qwen3 235B A22B Instruct 2507 by 26.152.979+26.1
- SimpleQAQwen3 235B A22B Instruct 2507 by 24.230.154.3+24.2
- τ²-Bench Telecom (AA run)Qwen3 235B A22B Instruct 2507 by 21.911.433.3+21.9
- ZebraLogicQwen3 235B A22B Instruct 2507 by 16.378.795+16.3
- AA-OmniscienceDeepSeek-R1 by 11.8-32.2-44+11.8
- IFEvalQwen3 235B A22B Instruct 2507 by 5.483.388.7+5.4
- ARC-AGI-1DeepSeek-R1 by 4.815.811+4.8
- XSTestQwen3 235B A22B Instruct 2507 by 4.294.498.6+4.2
- Aider-PolyglotQwen3 235B A22B Instruct 2507 by 453.357.3+4
- MMLU-ProXQwen3 235B A22B Instruct 2507 by 3.975.579.4+3.9
- anthropic_red_teamQwen3 235B A22B Instruct 2507 by 2.797.299.9+2.7
- Artificial Analysis Coding IndexDeepSeek-R1 by 2.524.622.1+2.5
- simple_safety_testsQwen3 235B A22B Instruct 2507 by 298100+2
- MMLU-ProDeepSeek-R1 by 18483+1
- AA Intelligencetie11.412tie
- bbqtie96.696.2tie
- AIME 2025tie7070.3tie
- MMLU-Reduxtie92.993.1tie
Questions people ask
Which is better, DeepSeek-R1 or Qwen3 235B A22B Instruct 2507?
Qwen3 235B A22B Instruct 2507 wins three of the six areas where both have results: agents, reasoning and following instructions. DeepSeek-R1 wins long documents. They are level on coding and facts.
Which is better for coding?
Neither. They win 1 coding test each of the 2 both models report.
Which is cheaper?
DeepSeek-R1 costs $2.00 per million input tokens and $4.00 per million output tokens; Qwen3 235B A22B Instruct 2507 costs $0.23 and $0.92. That makes Qwen3 235B A22B Instruct 2507 about 81% cheaper for the same work.
How do you compare the two?
We use the 29 benchmark tests both models have published scores on. The verdict counts the 10 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 19 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.