Qwen3 Max (Preview)
Sep 2025
Qwen3 Max (Preview)Compare inon
Benchmarks of Qwen3 Max (Preview)30
BenchmarkByScoreRank
Overall1
Knowledge4
Science2
Coding1
Writing16
Health & law1
Documents1
Legacy · saturated or no longer reported4
No benchmark matches
Rankingin non-thinking mode?
Alternatives?
ModelScoreCostPer taskWhat changes
CheaperMiMo v2.6 ProthinkingOpen×4.9-64%$0.13Pass back reasoning_content+3
BetterGPT-6 Lunalow+100%-87%$0.005Function calling only with none · Chat Completions+3
CheaperMiniMax M3.1 Flashthinking×4.7-83%$0.22Complete model response must be appended
CheaperStep 5 PreviewthinkingOpen×4.6-6%$1.03web_search not available · Chat Completions+2
CheaperClaude Haiku 5.5max×4.4-87%$0.21Remove sampling parameters+5ModelScoreCostPer taskWhat changes
MiMo v2.6 ProthinkingOpen×4.9-64%$0.13Pass back reasoning_content+3MiniMax M3.1 Flashthinking×4.7-83%$0.22Complete model response must be appendedStep 5 PreviewthinkingOpen×4.6-6%$1.03web_search not available · Chat Completions+2Claude Haiku 5.5max×4.4-87%$0.21Remove sampling parameters+5
DeepSeek V4.1 Flash 1017maxOpen×4.3-66%$0.26Pass back reasoning_content · Chat Completions+4ModelScoreCostPer taskWhat changes
GPT-6 Lunalow+100%-87%$0.005Function calling only with none · Chat Completions+3GPT-5.6 Lunalow+95%-70%$0.01MiMo v2.5thinkingOpen×2.7-89%$0.019Claude Haiku 5.5low×2.7-87%$0.024Remove sampling parameters+5
Mercury 2high+69%-75%$0.039Context: 262K → 128KModelScoreCostPer taskWhat changes
MiMo v2.6 ProthinkingOpen×4.9-64%$0.13Pass back reasoning_content+3
Kimi K3maxOpen×4.6+294%$2.00Sampling parameters are fixed+3Step 5 PreviewthinkingOpen×4.6-6%$1.03web_search not available · Chat Completions+2
Muse Spark OpenxhighOpen×4.3+31%$0.84Reasoning off leaves output unchanged · Self-hosted / llama.cpp+1DeepSeek V4.1 Flash 1017maxOpen×4.3-66%$0.26Pass back reasoning_content · Chat Completions+4ModelScoreCostPer taskWhat changes
Qwen3.8 Max (0902)thinking×4.2—$5.41Qwen3.8 2.4T A95BthinkingOpen×4.2—$2.16max_tokens range · QwenCloud+4Qwen3.7 Maxthinking×4—$1.15Qwen3.8 FlashthinkingOpen×3.7—$0.37max_tokens range · QwenCloud+5Qwen3.7 Plusthinking×3.5—$0.22Qwen3.6 Max Previewthinking×3—$1.30 / $7.80Include the word JSON · QwenCloudQwen3.5 397B A17BthinkingOpen×2.9—$0.47Qwen3 Max Thinkingthinking×2.8—$1.20 / $6Context: 262K → 256KAll Alibaba (Qwen) models →History
7 versionsBenchmarks
all categoriesChat & Knowledge
Humanity's Last Exam10.1#189
AA Omniscience Index-42.8#183
CritPt (physics)0.9#130
AA Intelligence Index 4.3.2≈12.6#175
AA Omniscience · Accuracy24.2#123
GPQA Diamond76.4#140
Coding
LiveCodeBench · AA archived65.1#63
Agents & Tool Use
τ²-bench · AA32.7#143
Writing & Language
IFBench (instructions)48#124
FAQ
How much does Qwen3 Max (Preview) cost?›
$0.87 per 1M input tokens and $3.48 per 1M output tokens (cheapest host: Qiniu).
Is Qwen3 Max (Preview) free to use?›
There is no free API; the cheapest paid price is $0.87 per 1M input tokens.
How good is Qwen3 Max (Preview), and what is it best at?›
#188 of 334 in Chat & Knowledge, #93 of 239 in Coding, #141 of 242 in Agents & Tool Use, #74 of 195 in Writing & Language.
Is Qwen3 Max (Preview) better than Qwen3 Coder Next?›
On Humanity's Last Exam, Qwen3 Max (Preview) scores 10.1 vs 10.1 — about the same.
How can I use Qwen3 Max (Preview)?›
Through the Alibaba (Qwen) API.
What is the context window of Qwen3 Max (Preview)?›
262K tokens.
When was Qwen3 Max (Preview) released? Is there a newer version?›
Released 2025-09-05. It has been superseded by Qwen3 Max.
Is Qwen3 Max (Preview) open source?›
No, Qwen3 Max (Preview) is a closed model available through an API. The closest open-weights option here is MiMo v2.6 Pro.
Can Qwen3 Max (Preview) read images?›
No, it accepts text only.
What is Qwen3 Max (Preview)?›
Qwen3 Max (Preview) is a closed language model from Alibaba (Qwen), released Sep 2025.
What are the best alternatives to Qwen3 Max (Preview)?›
MiMo v2.6 Pro (cheaper), GPT-6 Luna (better).