QwQ 32B Preview
Nov 2024
QwQ 32B PreviewCompare inon
Benchmarks of QwQ 32B Preview13
BenchmarkByScoreRank
Legacy · saturated or no longer reported2
No benchmark matches
Ranking?
Alternatives?
ModelScoreCostPer taskWhat changes
CheaperQwen 3.5 Flashunstatedwins 78%-16%$0.065 / $0.26
BetterGPT-5 Minihighwins 75%+409%$0.053
OpenGemma 4 31Bunstatedwins 82%+19%$0.10 / $0.34
CheaperNemotron 3 Nano 30B A3BunstatedOpenwins 75%-22%$0.06 / $0.24CheaperGemma 3 27BunstatedOpenwins 75%-26%$0.08 / $0.16ModelScoreCostPer taskWhat changes
Qwen 3.5 Flashunstatedwins 78%-16%$0.065 / $0.26Nemotron 3 Nano 30B A3BunstatedOpenwins 75%-22%$0.06 / $0.24Gemma 3 27BunstatedOpenwins 75%-26%$0.08 / $0.16GPT OSS 120BunstatedOpenwins 75%-56%$0.03 / $0.15Memory (4-bit): 20 GB → 70 GBGemma 3 12BunstatedOpenwins 72%-54%$0.05 / $0.10ModelScoreCostPer taskWhat changes
GPT-5 Minihighwins 75%+409%$0.053GPT-6 Lunamaxwins 77%+48%$0.068Function calling only with none · Chat Completions+3
Claude Haiku 5.5highwins 79%+48%$0.079Remove sampling parameters+5GPT-5.6 Lunaxhighwins 81%+233%$0.085GPT-5.6 Terraxhighwins 82%+3,233%$0.63Prompts with >272K input tokensModelScoreCostPer taskWhat changes
Gemma 4 31BunstatedOpenwins 82%+19%$0.10 / $0.34Qwen3.8 27BunstatedOpenwins 82%+608%$0.42 / $2.55max_tokens range · QwenCloud+4Gemma 4 26B A4BunstatedOpenwins 82%+94%$0.15 / $0.60Qwen 3.5 27BunstatedOpenwins 80%+511%$0.30 / $2.40Qwen3 Next 80B A3B InstructunstatedOpenwins 80%+548%$0.50 / $2Memory (4-bit): 20 GB → 49 GBModelScoreCostPer taskWhat changes
Qwen 3.7 Max Previewwins 84%—$2.50 / $7.50Context: not verifiedQwen3.7 Plusunstatedwins 83%—$0.32 / $1.28Qwen3.6 Max Previewunstatedwins 82%—$1.30 / $7.80Include the word JSON · QwenCloudQwen3.8 27BunstatedOpenwins 82%—$0.42 / $2.55max_tokens range · QwenCloud+4Qwen3.6 Plusunstatedwins 82%—$0.50 / $3Qwen3.5 397B A17BunstatedOpenwins 82%—$0.60 / $3.60Memory (4-bit): 20 GB → 242 GBQwen3 Max (Preview)unstatedwins 82%—$0.87 / $3.48Qwen3 235B A22B Thinking 2507unstatedOpenwins 80%—$0.23 / $2.30Memory (4-bit): 20 GB → 141 GB+1All 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 · overall1,154#189
Arena Text · Chinese1,218#170
Arena Text · English1,184.2#191
Arena Text · instruction following1,155.8#187
Arena Text · non-English1,129.5#190
Arena Text · Russian1,114.4#191
MMLU-Pro · official boardlab71
Arena Text · overall · October 22, 20261,161.8#185
Arena Text · overall · October 15, 20261,161.8#175
MathArena · AIME 202533.3#28
Writing & Language
Arena Text · creative writing1,104#188
FAQ
How much does QwQ 32B Preview cost?›
$0.12 per 1M input tokens and $0.18 per 1M output tokens (cheapest host: DeepInfra).
Is QwQ 32B Preview free to use?›
There is no free API; the cheapest paid price is $0.12 per 1M input tokens. The weights are open, so you can run it yourself.
How good is QwQ 32B Preview, and what is it best at?›
#191 of 205 in Chat & Knowledge, #183 of 197 in Writing & Language.
Is QwQ 32B Preview better than Mixtral 8x22B Instruct?›
On Arena Text · English, QwQ 32B Preview scores 1,184.2 vs 1,187.1 — about the same.
How can I use QwQ 32B Preview?›
Through the API of DeepInfra.
What is the context window of QwQ 32B Preview?›
33K tokens.
When was QwQ 32B Preview released? Is there a newer version?›
Released 2024-11-27. It is the current version in its line.
Is QwQ 32B Preview open source?›
Yes, the weights are open (Apache-2.0 licence).
Can I run QwQ 32B Preview locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can QwQ 32B Preview read images?›
No, it accepts text only.
What is QwQ 32B Preview?›
QwQ 32B Preview is an open-weights language model from Alibaba (Qwen), released Nov 2024.
What are the best alternatives to QwQ 32B Preview?›
Qwen 3.5 Flash (cheaper), GPT-5 Mini (better), Gemma 4 31B (open weights).