QWEN3.8 IS NO LONGER A PREVIEW — CHECKED 8 OCTOBER 2026
The Qwen3.8 family shipped in August 2026:
● Qwen3.8 flagship — 2.4T parameters, custom Qwen3.8-Max licence
● Qwen3.8-27B — Apache 2.0, around $0.80 / $4.00 per million on Groq
● Qwen3.8-Flash-Next — separate licence required for model-as-a-service
Qwen3.7-Max is the previous generation. Where the text below says "Preview", read it as the shipped release.
QUICK VERDICT — AUGUST 2026
Stay on Qwen3.7-Max: Production API at $2.50/$7.50/M, SWE-bench Verified 80.4%, stable endpoint. If Qwen3.7-Max meets your needs, there is no evidence yet that 3.8 is better on coding or reasoning tasks.
Test Qwen3.8-Max-Preview: If your workload needs multimodal input (images, video, documents) that 3.7 cannot handle, test 3.8 via Token Plan. Migrate production traffic only after Alibaba publishes a benchmark table and confirmed per-token pricing.
Last updated August 3, 2026. Related: Qwen3.8-Max full fact-check →