Are Qwen3.8-Max open weights out yet?

Alibaba announced that weights for Qwen3.8-Max and Qwen3.8-27B would be published on Hugging Face and ModelScope in the week of August 10, 2026 — making Qwen3.8-Max the first Max-class Qwen model ever released with open weights. The hosted API has been generally available since August 3, 2026 at $2 input / $6 output per million tokens.

How much does Qwen3.8-Max cost?

The Qwen3.8-Max hosted API is $2 input / $6 output per million tokens (flat), GA since August 3, 2026. That undercuts Kimi K3's $3 / $15. The open weights themselves are free to download but carry a custom license with conditions for some large model-as-a-service and consumer deployments.

Quick Answer

Qwen3.8-Max Open Weights: Released Week of Aug 10, 2026

Q: Is Qwen3.8-Max better than Kimi K3 and DeepSeek V4 Flash?

On benchmarks, Qwen3.8-Max performs closely to Kimi K3 and DeepSeek V4 Flash across most categories and is stronger at coding and software tasks. Artificial Analysis scored its max configuration at 57. For a hosted API, Qwen3.8-Max offers the cheaper cost-and-capability proposition; Kimi K3 offers a longer track record of released weights.

Published: August 11, 2026

The Short Answer

As of August 11, 2026, Alibaba is releasing Qwen3.8-Max open weights the week of August 10, 2026 on Hugging Face and ModelScope — the first Max-class Qwen model ever released open. The hosted API has been GA since August 3, 2026 at $2 in / $6 out per MTok, undercutting Kimi K3’s $3/$15.

What Is Qwen3.8-Max?

Qwen3.8-Max is Alibaba’s flagship open-weight model, sitting at the top of the Qwen 3.8 family. A smaller Qwen3.8-27B is dropping in the same window for people who want a model that fits on a single high-end GPU. This is a milestone: previous Max-class Qwen checkpoints stayed API-only, so open weights at this tier is new.

Pricing and License

Hosted API: $2 input / $6 output per MTok (flat), GA August 3, 2026.
Open weights: free to download, but under a custom license with conditions for some large model-as-a-service and consumer deployments — read it before you build a product on top.
Compare: Kimi K3 is $3/$15, DeepSeek V4 Flash is $0.14/$0.28 (off-peak).

How It Compares

On independent benchmarks, Qwen3.8-Max tracks closely with Kimi K3 and DeepSeek V4 Flash, with an edge on coding and software tasks. Artificial Analysis scored its max configuration at 57. For raw hosted cost-per-capability it’s the strongest of the three; Kimi K3 has the longer open-weights track record and simpler flat pricing.

How to Self-Host

Once weights land on Hugging Face / ModelScope, serve with vLLM or SGLang for throughput, or Ollama / LM Studio for local experimentation. Max-class means multi-GPU for the full model — reach for Qwen3.8-27B if you want single-GPU or Mac-M-series inference.

The Reality Check

“Open weights” here is weights-available under a custom license, not Apache-2.0 — verify the terms match your use case. For a fully permissive Chinese open model, DeepSeek V4 and Kimi K3 remain the safer defaults; Qwen3.8-Max wins on coding quality and hosted price.

Sources

Qubrid AI — Kimi K3 vs Qwen3.8-Max comparison (weights week of Aug 10): qubrid.com
OrcaRouter — Qwen 3.8 vs Kimi K3 (GA Aug 3, $2/$6): orcarouter.ai
Qwen — official model site: qwen.ai