AI agents · OpenClaw · self-hosting · automation

Quick Answer

Is Qwen3.8-Max Open Weights Out Yet? (Aug 5)

Published:

The Short Answer

No — Qwen3.8-Max open weights were not out yet as of August 5, 2026. Alibaba launched the model August 3 and said the weights would drop “the following week” (~week of Aug 10), alongside Qwen3.8-27B. The model is already usable via API/chat ahead of the weights.

Timeline

  • Aug 3, 2026 — Qwen3.8-Max launched; made widely accessible via API/chat; open weights promised “next week.”
  • ~Week of Aug 10, 2026 — expected open-weight release (Qwen3.8-Max + Qwen3.8-27B).
  • API pricing — TBD at launch.

What It Is

Qwen3.8-Max is Alibaba’s 2.4-trillion-parameter MoE model, tuned for agentic computer use, with a 1M context. It’s the largest open-weight model Alibaba has disclosed.

Benchmark Claims (vendor-reported)

BenchmarkScore
OSWorld-Verified86.1
Terminal-Bench 2.186.6
SWE-bench Pro67.7
FrontierSWE73.5
GPQA Diamond92.6
PaperBench93.0

Alibaba claims leads over GPT-5.6 Sol Max (83.2) and Fable 5 (85.0) on OSWorld. Independent confirmation was not available as of Aug 5, 2026, and a full model card lagged the launch.

What To Do Now

  • Want to test it → use the API/chat that went live Aug 3.
  • Planning to self-host → wait for the actual weights + license (~Aug 10) and size your GPUs for a 2.4T MoE.
  • Need proven open weights today → Kimi K3 (open weights Jul 27) or DeepSeek V4.

Sources