Compare / Qwen3-235B-A22B-Instruct-2507 vs Gemma 3 12B

Qwen3-235B-A22B-Instruct-2507 vs Gemma 3 12B

Which is cheaper, which is more capable, and which does better at each kind of work, from 24 results on 4 sources that measured both models.

AlibabaQwen3-235B-A22B-Instruct-2507 Released 21 Jul 2025 · $0.15 in · $0.75 out
GoogleGemma 3 12B Released 11 Mar 2025 · $0.050 in · $0.15 out
Qwen3-235B-A22B-Instruct-2507 wins 211 too close to callGemma 3 12B wins 2

Of the 24 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 Google, every model from both labs.

Qwen3-235B-A22B-Instruct-2507 or Gemma 3 12B? The short answer

Pick Qwen3-235B-A22B-Instruct-2507 if quality matters most: it scores higher (Intelligence Index 12.0 against 3.8). Pick Gemma 3 12B if cost matters more: it costs 4.0 times less per token.

  • Gemma 3 12B is cheaper: $0.075 against $0.30 per million tokens, 4.0 times less.
  • Qwen3-235B-A22B-Instruct-2507 scores higher on the Artificial Analysis Intelligence Index: 12.0 against 3.8.
  • Qwen3-235B-A22B-Instruct-2507 is newer: released 21 Jul 2025, against 11 Mar 2025 for Gemma 3 12B.
  • On third-party benchmarks, on the results both models have: Qwen3-235B-A22B-Instruct-2507 does better for agents and tool use, reasoning and knowledge and writing.

Head to head

12.0
Intelligence Index
3.8
$0.30
Price per million tokens, blendedlower is better
$0.075
262,144
Context window, tokens
131,072

Every fact side by side

Qwen3-235B-A22B-Instruct-2507Gemma 3 12B
Price per million tokens$0.15 in · $0.75 out$0.050 in · $0.15 out
Blended (3 in : 1 out)$0.30$0.075
Cheapest host, blended$0.15 (GMICloud, FP8)$0.075 (DeepInfra, BF16)
Intelligence Index12.03.8
Spring Prompt overall––
Output speed, tokens a second––
Context window262,144 tokens131,072 tokens
Released21 Jul 2025 (newer)11 Mar 2025
WeightsOpenOpen
Developer based inChinathe US

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.

Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 51 of 164 149 of 164
Reasoning and knowledgeHard questions across science, maths and general knowledge 129 of 198 198 of 198
WritingWhat people prefer in blind comparisons, and judged writing quality 85 of 188 149 of 188
Sticking to the factsSummarising without inventing things, and factual answers 98 of 152 5 of 152

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.

MeasureQwen3-235B-A22B-Instruct-2507Gemma 3 12B
Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results
Overall · TextArena rating, higher is better 109 · 204 of 387 measured 1,422 1,342
Business, management and finance · TextArena rating, higher is better 97 · 155 of 380 measured 1,432 1,386
Creative writing · TextArena rating, higher is better 122 · 173 of 385 measured 1,378 1,333
Artificial Analysis reported by Artificial Analysis · 3 results
Artificial Analysis Intelligence Indexindex score, higher is better 153 · 243 of 243 measured 12.0 3.8
GPQA Diamond% of questions, higher is better 136 · 224 of 231 measured 75.3% 34.9%
Humanity's Last Exam% of questions, higher is better 144 · 213 of 242 measured 11.1% 4.2%
Berkeley Function Calling Leaderboard (BFCL) V4 reported by Berkeley Function Calling Leaderboard (BFCL) V4 · 3 results
Overall accuracy · BFCL% correct, higher is better 20 · 48 of 81 measured 52.2%bfcl prompt 30.4%bfcl prompt
Irrelevance detection · BFCL% correct, higher is better 41 · 62 of 81 measured 81.7%bfcl fc 70.3%bfcl prompt
Memory · BFCL% correct, higher is better 26 · 18 of 81 measured 23.9%bfcl fc 27.5%bfcl prompt
UGI Leaderboard reported by UGI Leaderboard · 3 results
Writing score · UGIscore out of 100, higher is better 80 · 156 of 230 measured 49.0 29.9
Requested-length error · UGI% off the requested word count, lower is better 69 · 33 of 230 measured 15.0% 9.0%
Style adherence · UGIscore from 0 to 1, higher is better 64 · 212 of 230 measured 0.36 0.29
Show all 24 results (12 more)
MeasureQwen3-235B-A22B-Instruct-2507Gemma 3 12B
Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results
Expert prompts · TextArena rating, higher is better 103 · 257 of 338 measured 1,447 1,274
Instruction following · TextArena rating, higher is better 104 · 207 of 387 measured 1,414 1,321
Writing, literature and language · TextArena rating, higher is better 115 · 208 of 386 measured 1,396 1,313
Artificial Analysis reported by Artificial Analysis · 4 results
IFBench% of instructions, higher is better 113 · 151 of 195 measured 46.1% 36.7%
Long-context reasoning (AA-LCR)% of questions, higher is better 177 · 216 of 233 measured 33.9% 8.3%
Terminal-Bench Hard% of tasks, higher is better 110 · 180 of 190 measured 15.2% 0.8%
τ²-bench telecom% of tasks, higher is better 127 · 183 of 193 measured 33.3% 10.8%
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 · 60 of 81 measured 45.4%bfcl fc 5.8%bfcl prompt
Relevance detection · BFCL% correct, higher is better 8 · 8 of 81 measured 93.8%bfcl prompt 93.8%bfcl prompt
Single-turn calls (curated) · BFCL% correct, higher is better 2 · 63 of 81 measured 90.3%bfcl prompt 79.4%bfcl prompt
Single-turn calls (user-contributed) · BFCL% correct, higher is better 15 · 37 of 81 measured 78.7%bfcl prompt 74.2%bfcl prompt
Web search · BFCL% correct, higher is better 20 · 46 of 81 measured 54.0%bfcl fc 4.0%bfcl prompt

Quick answers

Is Qwen3-235B-A22B-Instruct-2507 better than Gemma 3 12B?

Pick Qwen3-235B-A22B-Instruct-2507 if quality matters most: it scores higher (Intelligence Index 12.0 against 3.8). Pick Gemma 3 12B if cost matters more: it costs 4.0 times less per token.

Which is cheaper, Qwen3-235B-A22B-Instruct-2507 or Gemma 3 12B?

Gemma 3 12B is cheaper: $0.075 against $0.30 per million tokens, 4.0 times less.

Which is better for agents and tool use, Qwen3-235B-A22B-Instruct-2507 or Gemma 3 12B?

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 149 for Gemma 3 12B.

Which is better for reasoning and knowledge, Qwen3-235B-A22B-Instruct-2507 or Gemma 3 12B?

Qwen3-235B-A22B-Instruct-2507 ranks 129 of 198 models on sale for hard questions across science, maths and general knowledge, against 198 for Gemma 3 12B.

Which is better for writing, Qwen3-235B-A22B-Instruct-2507 or Gemma 3 12B?

Qwen3-235B-A22B-Instruct-2507 ranks 85 of 188 models on sale for what people prefer in blind comparisons, and judged writing quality, against 149 for Gemma 3 12B.

Which is better for sticking to the facts, Qwen3-235B-A22B-Instruct-2507 or Gemma 3 12B?

Gemma 3 12B ranks 5 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 Gemma 3 12B?

Qwen3-235B-A22B-Instruct-2507: 262,144 tokens, against 131,072 for Gemma 3 12B.

Run Qwen3-235B-A22B-Instruct-2507 and Gemma 3 12B on your own prompt

Benchmarks aren't your data. Try both side by side in the playground, with the cost of every answer.

Try the playgroundTalk to us