Model comparison
DeepSeek V4 Flash vs Qwen3 235B-A22B
DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 43.2 on the Noometry Index.
Last verified . 33 shared benchmarks.
DeepSeek
Alibaba (Qwen)Summary
- They share 33 benchmarks with published results for both. DeepSeek V4 Flash scores higher in 7 categories and Qwen3 235B-A22B in 1 category; 7 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek V4 Flash leads 53.7 to 15.8.
- The biggest single-benchmark swing is ARC-AGI-1: 89% for DeepSeek V4 Flash and 11% for Qwen3 235B-A22B.
- DeepSeek V4 Flash is cheaper at $0.15 / $0.60 per million input/output tokens, against $0.70 / $2.80 for Qwen3 235B-A22B.
- DeepSeek V4 Flash accepts more context: 1M tokens versus 131K.
Side by side
| DeepSeek V4 Flash | Qwen3 235B-A22B | |
|---|---|---|
| Provider | DeepSeek | Alibaba (Qwen) |
| Noometry Index | 53.6 | 43.2 |
| Released | 2026-04-24 | 2025-04 |
| Weights | Open | Open |
| Context window | 1M | 131K |
| Max output | 393K | 16K |
| Input $ / M tokens | $0.15 | $0.70 |
| Output $ / M tokens | $0.60 | $2.80 |
| Results tracked | 41 | 49 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Flash leads
DeepSeek V4 Flash: 47.6 (#56), Qwen3 235B-A22B: 43.8 (#72)
| Benchmark | DeepSeek V4 Flash | Qwen3 235B-A22B |
|---|---|---|
| SciCode | 49.9% | 42.4% |
| WeirdML | 63% | 41% |
| LMArena Coding | 1457 | 1445 |
| FrontierCode | 18.8% | — |
| Aider Polyglot | — | 59.6% |
| LMArena WebDev | 1581 | — |
| ALE-Bench | 1,306 | — |
Agentic & Tool Use Not comparable
DeepSeek V4 Flash: —, Qwen3 235B-A22B: 33.7 (#48)
| Benchmark | DeepSeek V4 Flash | Qwen3 235B-A22B |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 52.1% |
| Vending-Bench 2 | — | -11.34 |
Reasoning DeepSeek V4 Flash leads
DeepSeek V4 Flash: 53.7 (#25), Qwen3 235B-A22B: 15.8 (#302)
| Benchmark | DeepSeek V4 Flash | Qwen3 235B-A22B |
|---|---|---|
| ARC-AGI-2 | 61.4% | 1.3% |
| SimpleBench | 61.1% | 31% |
| Kagi LLM Benchmark | 52.2% | 69.4% |
| ARC-AGI-1 | 89% | 11% |
| CritPt | 16.6% | 0% |
| Chess Puzzles | 33% | 12% |
| LMArena Hard Prompts | 1443 | 1433 |
| Mystery Game Puzzles | 34% | 9% |
| DTBench | 90.9% | 80.3% |
| LMCA | 41.7% | 29.3% |
| Epoch Capabilities Index | 154.51 | 143.87 |
| NYT Connections (extended) | 89.6% | — |
| ForecastBench | — | 59.7 |
Math DeepSeek V4 Flash leads
DeepSeek V4 Flash: 60.0 (#31), Qwen3 235B-A22B: 48.5 (#51)
| Benchmark | DeepSeek V4 Flash | Qwen3 235B-A22B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 94.4% | 86.7% |
| LMArena Math | 1427 | 1431 |
| FrontierMath (Tiers 1-3) | 57.5% | — |
| FrontierMath Tier 4 | 24.4% | — |
| MathArena Final-Answer Competitions | 76.5% | — |
| ProofBench | 56% | — |
| Omni-MATH | — | 71.8% |
| MATH Level 5 | — | 68.9% |
| FrontierMath (Feb 2025 set) | — | 8.5% |
| FrontierMath Tier 4 (v1) | — | 0% |
Knowledge DeepSeek V4 Flash leads
DeepSeek V4 Flash: 55.5 (#43), Qwen3 235B-A22B: 49.5 (#67)
| Benchmark | DeepSeek V4 Flash | Qwen3 235B-A22B |
|---|---|---|
| GPQA Diamond | 91% | 80.1% |
| SimpleQA Verified | 33.6% | 40.4% |
| LMArena Expert | 1441 | 1460 |
| MMLU-Pro | — | 84.4% |
| Confabulations | — | 15.6% |
| Vectara Hallucination Rate | — | 9.3% |
| GPQA (HELM) | — | 72.7% |
Multilingual Too close to call
DeepSeek V4 Flash: 53.2 (#71), Qwen3 235B-A22B: 52.5 (#84)
| Benchmark | DeepSeek V4 Flash | Qwen3 235B-A22B |
|---|---|---|
| LMArena Non-English | 1420 | 1410 |
| LMArena Chinese | 1467 | 1481 |
| LMArena French | 1438 | 1445 |
| LMArena German | 1418 | 1432 |
| LMArena Japanese | 1403 | 1400 |
| LMArena Korean | 1383 | 1391 |
| LMArena Russian | 1428 | 1411 |
| LMArena Spanish | 1437 | 1429 |
Instruction Following DeepSeek V4 Flash leads
DeepSeek V4 Flash: 75.1 (#76), Qwen3 235B-A22B: 72.8 (#132)
| Benchmark | DeepSeek V4 Flash | Qwen3 235B-A22B |
|---|---|---|
| LMArena Instruction Following | 1421 | 1408 |
| IFEval | — | 83.5% |
Long Context Qwen3 235B-A22B leads
DeepSeek V4 Flash: 43.8 (#81), Qwen3 235B-A22B: 46.1 (#25)
| Benchmark | DeepSeek V4 Flash | Qwen3 235B-A22B |
|---|---|---|
| LMArena Longer Query | 1435 | 1426 |
| Fiction.LiveBench | — | 75% |
Writing & Preference DeepSeek V4 Flash leads
DeepSeek V4 Flash: 64.0 (#59), Qwen3 235B-A22B: 59.8 (#103)
| Benchmark | DeepSeek V4 Flash | Qwen3 235B-A22B |
|---|---|---|
| LMArena Text | 1432 | 1419 |
| LMArena Creative Writing | 1403 | 1384 |
| EQ-Bench Creative Writing | 1559 | 1366 |
| LMArena Multi-Turn | 1449 | 1432 |
| Short-Story Creative Writing | — | 83% |
| WildBench | — | 86.6% |
Frequently asked questions
Is DeepSeek V4 Flash better than Qwen3 235B-A22B?
DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 43.2 on the Noometry Index.
Which is cheaper, DeepSeek V4 Flash or Qwen3 235B-A22B?
DeepSeek V4 Flash is cheaper. It lists at $0.15 per million input tokens and $0.60 per million output tokens; Qwen3 235B-A22B lists at $0.70 and $2.80.
Is DeepSeek V4 Flash or Qwen3 235B-A22B better for coding?
DeepSeek V4 Flash scores higher on coding benchmarks: 47.6 versus 43.8 in the Noometry coding category.
Which has the bigger context window?
DeepSeek V4 Flash does, with 1M tokens against 131K.
How many benchmarks do DeepSeek V4 Flash and Qwen3 235B-A22B share?
33 benchmarks have published results for both models. DeepSeek V4 Flash has 41 scored results on Noometry and Qwen3 235B-A22B has 49.