Qwen3.8 2.4T A95B API
Qwen3.8-2.4T-A95B is the open-source release of Alibaba's large Qwen3.8 model: 2.4T total parameters, about 95B active, a 1M context and 131K output. The official page reports GPQA Diamond 92.6, PaperBench 93.0, OSWorld 86.1, BabyVision 82.0 and fourth place on CodeArena.
Official source: model page · verified 2026-10-22
Pricing
Prices can change with upstream providers; the rate at settlement is authoritative.
Call the API
Use the OpenAI-compatible endpoint; no model-specific SDK setup is required.
curl https://api.kimiseek.app/v1/chat/completions \
-H "Authorization: Bearer $KIMISEEK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen3.8-2.4t-a95b",
"messages": [{"role": "user", "content": "Hello from KimiSeek"}]
}'Where it fits
Best for
- Difficult reasoning and agent planning
- Coding and computer-use evaluation
- Teams that need an open-source large Qwen variant
Tradeoffs
- Premium pricing and heavy serving requirements
- The 131K output ceiling may require decomposition for very long deliverables
Model details
FAQ
What does A95B mean?
It refers to the sparse MoE variant with about 95B parameters activated per step, alongside 2.4T total parameters.
Is this the open-source Qwen3.8 release?
Yes. The official Model Studio page describes it as the open-source release of Qwen's latest flagship.
How do I call this model?
Use KimiSeek's OpenAI-compatible API and set the model field to qwen3.8-2.4t-a95b with your KimiSeek API key.