Compare / Qwen3-235B-A22B-Instruct-2507 vs DeepSeek-V3.2-Exp
Qwen3-235B-A22B-Instruct-2507 vs DeepSeek-V3.2-Exp
Which is cheaper, which is more capable, and which does better at each kind of work, from 25 results on 4 sources that measured both models.
Of the 25 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.
Results as of 9 October 2026 (catalogue release 2026-10-09-dd620e778ac5). See also Alibaba vs DeepSeek, every model from both labs.
Qwen3-235B-A22B-Instruct-2507 or DeepSeek-V3.2-Exp? The short answer
Pick DeepSeek-V3.2-Exp: it scores higher (Intelligence Index 16.6 against 12.0) at about the same price.
- They cost about the same: $0.30 and $0.30 per million tokens (three input to one output).
- DeepSeek-V3.2-Exp scores higher on the Artificial Analysis Intelligence Index: 16.6 against 12.0.
- DeepSeek-V3.2-Exp is newer: released 29 Sep 2025, against 21 Jul 2025 for Qwen3-235B-A22B-Instruct-2507.
- On third-party benchmarks, on the results both models have: DeepSeek-V3.2-Exp does better for agents and tool use, reasoning and knowledge, writing and sticking to the facts.
Head to head
Every fact side by side
| Qwen3-235B-A22B-Instruct-2507 | DeepSeek-V3.2-Exp | |
|---|---|---|
| Price per million tokens | $0.15 in · $0.75 out | $0.27 in · $0.41 out |
| Blended (3 in : 1 out) | $0.30 | $0.30 |
| Cheapest host, blended | $0.15 (GMICloud, FP8) | – |
| Intelligence Index | 12.0 | 16.6 |
| Spring Prompt overall | – | – |
| Output speed, tokens a second | – | – |
| Context window | 262,144 tokens | 163,840 tokens |
| Released | 21 Jul 2025 | 29 Sep 2025 (newer) |
| Weights | Open | Open |
| Developer based in | China | China |
Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.
Which is better for what
Each model's rank among the models on sale today, for each kind of work, from our own benchmarks and licensed sources. Further right is better; the better rank is in green.
Every shared result
Each model's best result on every measure both have, grouped by benchmark, with the setting that produced it. Under each measure, every model that benchmark has measured is a grey tick, best to the right, so you can see where the two sit in the field. The headline measures of each benchmark come first; the rest are under "Show all". Green marks the better value; where a source reports ranges that overlap, neither is marked.
| Measure | Qwen3-235B-A22B-Instruct-2507 | DeepSeek-V3.2-Exp |
|---|---|---|
| Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results | ||
| Overall · TextArena rating, higher is better 109 · 105 of 387 measured | 1,422 | 1,425thinking reasoning |
| Overall · Text factualityArena rating, higher is better 94 · 66 of 163 measured | 1,431 | 1,446thinking reasoning |
| Business, management and finance · TextArena rating, higher is better 97 · 104 of 380 measured | 1,432 | 1,422 |
| Artificial Analysis reported by Artificial Analysis · 3 results | ||
| Artificial Analysis Intelligence Indexindex score, higher is better 153 · 119 of 243 measured | 12.0 | 16.6thinking reasoning |
| GPQA Diamond% of questions, higher is better 136 · 110 of 231 measured | 75.3% | 79.7%thinking reasoning |
| Humanity's Last Exam% of questions, higher is better 144 · 126 of 242 measured | 11.1% | 14.9%thinking reasoning |
| Berkeley Function Calling Leaderboard (BFCL) V4 reported by Berkeley Function Calling Leaderboard (BFCL) V4 · 3 results | ||
| Overall accuracy · BFCL% correct, higher is better 20 · 12 of 81 measured | 52.2%bfcl prompt | 56.7%thinking reasoning, bfcl prompt |
| Irrelevance detection · BFCL% correct, higher is better 41 · 8 of 81 measured | 81.7%bfcl fc | 93.2%bfcl fc |
| Memory · BFCL% correct, higher is better 26 · 7 of 81 measured | 23.9%bfcl fc | 54.2%bfcl fc |
| UGI Leaderboard reported by UGI Leaderboard · 3 results | ||
| Writing score · UGIscore out of 100, higher is better 80 · 64 of 230 measured | 49.0 | 54.0thinking reasoning |
| Requested-length error · UGI% off the requested word count, lower is better 69 · 147 of 230 measured | 15.0% | 25.0%thinking reasoning |
| Style adherence · UGIscore from 0 to 1, higher is better 64 · 113 of 230 measured | 0.36 | 0.35no reasoning |
Show all 25 results (13 more)
| Measure | Qwen3-235B-A22B-Instruct-2507 | DeepSeek-V3.2-Exp |
|---|---|---|
| Arena (formerly LMArena) reported by Arena (formerly LMArena) · 4 results | ||
| Creative writing · TextArena rating, higher is better 122 · 79 of 385 measured | 1,378 | 1,410 |
| Expert prompts · TextArena rating, higher is better 103 · 99 of 338 measured | 1,447 | 1,449thinking reasoning |
| Instruction following · TextArena rating, higher is better 104 · 98 of 387 measured | 1,414 | 1,416thinking reasoning |
| Writing, literature and language · TextArena rating, higher is better 115 · 90 of 386 measured | 1,396 | 1,411 |
| Artificial Analysis reported by Artificial Analysis · 4 results | ||
| IFBench% of instructions, higher is better 113 · 93 of 195 measured | 46.1% | 54.2%thinking reasoning |
| Long-context reasoning (AA-LCR)% of questions, higher is better 177 · 96 of 233 measured | 33.9% | 72.3%thinking reasoning |
| Terminal-Bench Hard% of tasks, higher is better 110 · 65 of 190 measured | 15.2% | 31.1%thinking reasoning |
| τ²-bench telecom% of tasks, higher is better 127 · 126 of 193 measured | 33.3% | 33.9%no reasoning |
| Berkeley Function Calling Leaderboard (BFCL) V4 reported by Berkeley Function Calling Leaderboard (BFCL) V4 · 5 results | ||
| Multi-turn tasks · BFCL% correct, higher is better 20 · 21 of 81 measured | 45.4%bfcl fc | 44.9%thinking reasoning, bfcl prompt |
| Relevance detection · BFCL% correct, higher is better 8 · 8 of 81 measured | 93.8%bfcl prompt | 93.8%thinking reasoning, bfcl prompt |
| Single-turn calls (curated) · BFCL% correct, higher is better 2 · 33 of 81 measured | 90.3%bfcl prompt | 85.5%thinking reasoning, bfcl prompt |
| Single-turn calls (user-contributed) · BFCL% correct, higher is better 15 · 29 of 81 measured | 78.7%bfcl prompt | 76.0%thinking reasoning, bfcl prompt |
| Web search · BFCL% correct, higher is better 20 · 14 of 81 measured | 54.0%bfcl fc | 69.5%bfcl fc |
Compare them with other models
Quick answers
Is Qwen3-235B-A22B-Instruct-2507 better than DeepSeek-V3.2-Exp?
Pick DeepSeek-V3.2-Exp: it scores higher (Intelligence Index 16.6 against 12.0) at about the same price.
Which is cheaper, Qwen3-235B-A22B-Instruct-2507 or DeepSeek-V3.2-Exp?
They cost about the same: $0.30 and $0.30 per million tokens (three input to one output).
Which is better for agents and tool use, Qwen3-235B-A22B-Instruct-2507 or DeepSeek-V3.2-Exp?
Qwen3-235B-A22B-Instruct-2507 ranks 51 of 164 models on sale for multi-step tasks with tools: support desks, coding agents, function calls, against 72 for DeepSeek-V3.2-Exp.
Which is better for reasoning and knowledge, Qwen3-235B-A22B-Instruct-2507 or DeepSeek-V3.2-Exp?
DeepSeek-V3.2-Exp ranks 106 of 198 models on sale for hard questions across science, maths and general knowledge, against 129 for Qwen3-235B-A22B-Instruct-2507.
Which is better for writing, Qwen3-235B-A22B-Instruct-2507 or DeepSeek-V3.2-Exp?
DeepSeek-V3.2-Exp ranks 71 of 188 models on sale for what people prefer in blind comparisons, and judged writing quality, against 85 for Qwen3-235B-A22B-Instruct-2507.
Which is better for sticking to the facts, Qwen3-235B-A22B-Instruct-2507 or DeepSeek-V3.2-Exp?
DeepSeek-V3.2-Exp ranks 40 of 152 models on sale for summarising without inventing things, and factual answers, against 98 for Qwen3-235B-A22B-Instruct-2507.
Which has the bigger context window, Qwen3-235B-A22B-Instruct-2507 or DeepSeek-V3.2-Exp?
Qwen3-235B-A22B-Instruct-2507: 262,144 tokens, against 163,840 for DeepSeek-V3.2-Exp.
Run Qwen3-235B-A22B-Instruct-2507 and DeepSeek-V3.2-Exp on your own prompt
Benchmarks aren't your data. Try both side by side in the playground, with the cost of every answer.