Qwen3.8 2.4T A95B
Qwen3.8 2.4T A95BRankingin thinking mode?
Full ranking →Alternatives?
CheaperGrok 4.6high+3%-31%$1.48
OpenGLM 5.3max+3%-7%$2.01Reasoning cannot be disabled+3
CheaperMuse Spark 1.3max+3%-26%$1.60Only auto tool_choice is supported+5CheaperQwen3.8 27BxhighOpen-2%-53%$1.01max_tokens range · QwenCloud+4CheaperGLM 5.3 FlashmaxOpen-4%-87%$0.28Defaults to max · Self-hosted+4Grok 4.6high+3%-31%$1.48Muse Spark 1.3max+3%-26%$1.60Only auto tool_choice is supported+5GLM 5.3maxOpen+3%-7%$2.01Reasoning cannot be disabled+3Qwen3.8 27BxhighOpen-2%-53%$1.01max_tokens range · QwenCloud+4GLM 5.3 FlashmaxOpen-4%-87%$0.28Defaults to max · Self-hosted+4GLM 5.3maxOpen+3%-7%$2.01Reasoning cannot be disabled+3Qwen3.8 27BxhighOpen-2%-53%$1.01max_tokens range · QwenCloud+4GLM 5.3 FlashmaxOpen-4%-87%$0.28Defaults to max · Self-hosted+4
Kimi K3maxOpen-6%-7%$2.00Memory (4-bit): 1,468 GB → 1,668 GB+4Qwen3.8 FlashthinkingOpen-8%-83%$0.37max_tokens range · QwenCloud+5Qwen3.8 Max (0902)thinking-3%+151%$5.41Qwen3.8 27BmediumOpen-3%-47%$1.13max_tokens range · QwenCloud+4Qwen3.8 FlashthinkingOpen-8%-83%$0.37max_tokens range · QwenCloud+5Qwen3.6 Plusthinking-58%—$0.50 / $3Qwen3.7 Plusthinking-64%-90%$0.22Qwen3.6 27BthinkingOpen-66%-71%$0.62Qwen 3.5 122B A10BthinkingOpen-69%-85%$0.32Qwen3.5 397B A17BthinkingOpen-73%-78%$0.47All Alibaba (Qwen) models →Run it yourself
estimateBenchmarks
in Agents & Tool UseWatch out 5
max_tokens range · QwenCloud›
With thinking enabled, max_tokens outside 1–32768 returns HTTP 400; use max_completion_tokens to budget reasoning plus answer without that cap. Source ↗
Thinking cannot be disabled · Self-hosted›
The 2.4T model requires thinking for every interaction and cannot disable it. Source ↗
tool_choice only supports auto or none · QwenCloud›
With thinking enabled, tool_choice supports only auto or none. Source ↗
Include the word JSON · QwenCloud›
JSON Object mode requires the word JSON in a system or user message; otherwise the API returns an error. Source ↗
Agent search_strategy not supported · QwenCloud / Chat Completions›
Chat Completions does not support Qwen3.8 agent search_strategy; use the Responses web_search tool for multi-turn retrieval. Source ↗
FAQ
How much does Qwen3.8 2.4T A95B cost?›
$2 per 1M input tokens and $6 per 1M output tokens; cached input $0.25 per 1M; cheapest via Featherless at $1 / $4. A typical benchmark task costs about $2.16 at default effort.
Is Qwen3.8 2.4T A95B free to use?›
There is no free API; the cheapest paid price is $1 per 1M input tokens. The weights are open, so you can run it yourself.
How good is Qwen3.8 2.4T A95B, and what is it best at?›
#35 of 334 in Chat & Knowledge, #38 of 132 in Coding, #26 of 133 in Agents & Tool Use, #18 of 91 in Writing & Language, #37 of 133 in Vision & Documents. Strongest at: Agents & work tasks, Reasoning, Science, Long context, Knowledge.
Is Qwen3.8 2.4T A95B better than GLM 5.3?›
On τ³-Banking · AA, Qwen3.8 2.4T A95B scores 49.1 vs 50.3 — lower. It is 7% more expensive per task.
How can I use Qwen3.8 2.4T A95B?›
Through the API of Alibaba, DeepInfra, Featherless, Novita, SiliconFlow and 1 more.
What is the context window of Qwen3.8 2.4T A95B?›
262K tokens, with up to 131K output tokens.
When was Qwen3.8 2.4T A95B released? Is there a newer version?›
Released 2026-08-12. It is the current version in its line.
Is Qwen3.8 2.4T A95B open source?›
Yes, the weights are open (qwen3.8-max licence).
Can I run Qwen3.8 2.4T A95B locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Qwen3.8 2.4T A95B read images?›
No, it accepts text only.
What is Qwen3.8 2.4T A95B?›
Qwen3.8 2.4T A95B is an open-weights language model from Alibaba (Qwen), released Aug 2026.
What are the best alternatives to Qwen3.8 2.4T A95B?›
Grok 4.6 (cheaper), GLM 5.3 (open weights).