QwQ 32B
Mar 2025
QwQ 32BCompare inon
Benchmarks of QwQ 32B25
BenchmarkByScoreRank
Knowledge1
Math3
Writing17
Legacy · saturated or no longer reported4
No benchmark matches
Ranking?
Alternatives?
ModelScoreCostPer taskWhat changes
CheaperGPT OSS 120BhighOpen×2-72%$0.092Memory (4-bit): 20 GB → 70 GBOpenQwen3 32Bunstated+91%+476%$0.70 / $2.80Context: 131K → 33KCheaperQwen2.5 Coder 32B InstructunstatedOpen-22%-61%$0.06 / $0.15BetterGPT-5high×4.2+1,518%$1.25 / $10Bettero3 Prohigh×4.1+16,371%$20 / $80ModelScoreCostPer taskWhat changes
GPT OSS 120BhighOpen×2-72%$0.092Memory (4-bit): 20 GB → 70 GBQwen2.5 Coder 32B InstructunstatedOpen-22%-61%$0.06 / $0.15ModelScoreCostPer taskWhat changes
GPT OSS 120BhighOpen×2-72%$0.092Memory (4-bit): 20 GB → 70 GBGPT-5high×4.2+1,518%$1.25 / $10o3 Prohigh×4.1+16,371%$20 / $80o3high×3.9+1,547%$2 / $8
R1 0528unstatedOpen×3.4+227%$0.39 / $1.63Context: 131K → 128K+1ModelScoreCostPer taskWhat changes
Qwen3 32BunstatedOpen+91%+476%$0.70 / $2.80Context: 131K → 33KQwen2.5 Coder 32B InstructunstatedOpen-22%-61%$0.06 / $0.15
Gemma 3 27BunstatedOpen-77%-53%$0.08 / $0.16ModelScoreCostPer taskWhat changes
Qwen3 32BunstatedOpen+91%—$0.70 / $2.80Context: 131K → 33KQwen2.5 Coder 32B InstructunstatedOpen-22%—$0.06 / $0.15Qwen3.8 Max (0902)thinkingn/a—$5.41Qwen3.8 FlashthinkingOpenn/a—$0.37Memory (4-bit): 20 GB → 108 GB+6Qwen3.8 27BmediumOpenn/a—$1.13max_tokens range · QwenCloud+4Qwen3.8 2.4T A95BthinkingOpenn/a—$2.16Memory (4-bit): 20 GB → 1,468 GB+5Qwen 3.7 Flash 2026-07-15n/a—$0.03 / $0.13Context: not verifiedQwen3.7 Flashunstatedn/a—$0.03 / $0.13All Alibaba (Qwen) models →Run it yourself
estimateFits on✗ 8 GB (phone, laptop)✗ 16 GB✓ 24 GB GPU · compressed✓ 48 GB GPU · 8-bit✓ 80 GB GPU · full quality
LicenceApache-2.0free for commercial use
Benchmarks
all categoriesChat & Knowledge
Arena Text · overall1,336#136
Arena Text · Chinese1,378.1#117
Arena Text · English1,350.6#131
Arena Text · French1,336.4#112
Arena Text · German1,312.8#109
Arena Text · instruction following1,296.9#136
Arena Text · Japanese1,262#97
Arena Text · Korean1,278.6#102
Arena Text · non-English1,305.4#131
Arena Text · Polish1,287.9#114
Arena Text · Russian1,297.1#131
Arena Text · Spanish1,353.4#112
EQ-Bench 31,167.7#32
GPQA Diamond pass@1lab54.5
AIME 2024 pass@1lab44
AIME 2024 cons@64lab60
MATH-500 pass@1lab90.6
Arena Text · overall · full cohort1,329.1#132
GPQA Diamond · Epoch65.3#88
OTIS Mock AIME 2024 202559.2#86
MathArena · AIME 202565.8#26
Coding
LiveCodeBench pass@1lab41.9
Codeforces Ratinglab1,316
Aider Polyglot · two attempts20.9#11
Writing & Language
Arena Text · creative writing1,293#134
EQ-Bench Creative Writing v31,257.2#70
Arena Text · creative writing · full cohort1,287.5#128
FAQ
How much does QwQ 32B cost?›
$0.15 per 1M input tokens and $0.40 per 1M output tokens (cheapest host: DeepInfra).
Is QwQ 32B free to use?›
There is no free API; the cheapest paid price is $0.15 per 1M input tokens. The weights are open, so you can run it yourself.
How good is QwQ 32B, and what is it best at?›
#136 of 196 in Chat & Knowledge, #11 of 15 in Coding, #134 of 195 in Writing & Language.
Is QwQ 32B better than GPT-4o?›
On Aider Polyglot · two attempts, QwQ 32B scores 20.9 vs 23.1 — lower.
How can I use QwQ 32B?›
Through the API of DeepInfra.
What is the context window of QwQ 32B?›
131K tokens.
When was QwQ 32B released? Is there a newer version?›
Released 2025-03-05. It is the current version in its line.
Is QwQ 32B open source?›
Yes, the weights are open (Apache-2.0 licence).
Can I run QwQ 32B locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can QwQ 32B read images?›
No, it accepts text only.
What is QwQ 32B?›
QwQ 32B is an open-weights language model from Alibaba (Qwen), released Mar 2025.
What are the best alternatives to QwQ 32B?›
GPT OSS 120B (cheaper), Qwen3 32B (open weights).