Qwen3.8-Max puts 2.4T parameters and open weights on the coding map
Original: Qwen3.8-Max puts 2.4T parameters and open weights on the coding map View original →
A 2.4T model moves into the open-weight lane
The competition around coding models is shifting from leaderboard placement to deployment control and cost. Qwen posted on X on August 3, 2026 that “open weights of Qwen3.8-Max will be released,” adding that Qwen3.8-27B is also moving to open weights. The tweet matters because Qwen is pairing an API launch for a 2.4T-parameter MoE flagship with a near-term weights release.
The QwenCloud model page describes Qwen3.8-Max as a 2.4T MoE model for coding and professional work. It lists a 1M-token context window, 991K maximum input, and 131K maximum output. The public pricing is $2 per million input tokens, $6 per million output tokens, and $0.25 per million tokens for implicit cache input. The page also lists built-in tools including code interpreter, web search, web extractor, and image search capabilities.
The larger claim is long-horizon autonomy. Qwen says the model can handle more than 10 days of self-evolving development from an empty folder to production, and points to the qwen-code-dev-bot/oh-my-cli GitHub repository as a project trace. That repository is an Apache-2.0 Node.js 22 and TypeScript code-agent CLI with file and shell tools. The tweet also cites 500-plus turns of chip design optimization and 365 days of e-commerce strategy as examples of extended planning.
The next check is external validation. The weights still need to arrive with clear licensing, and long-context coding claims need reproducible tests that measure not only task success but cost, latency, tool safety, and regression rates. If Qwen can make the open-weight release usable at the advertised scale, closed coding models will face a sharper price-and-control comparison. The source tweet is available on X.
Related Articles
A LocalLLaMA release post presents OmniCoder-9B as a Qwen3.5-9B-based coding agent fine-tuned on 425,000-plus agentic trajectories, with commenters focusing on its read-before-write behavior and usefulness at small model size.
HN latched onto the open-weight angle: a 35B MoE model with only 3B active parameters is interesting if it can actually carry coding-agent work. Qwen says Qwen3.6-35B-A3B improves sharply over Qwen3.5-35B-A3B, while commenters immediately moved to GGUF builds, Mac memory limits, and whether open-model-only benchmark tables are enough context.
LocalLLaMA reacted like dense models had suddenly become fun again. The official Qwen numbers were strong, but the real community energy came from people immediately asking about quants, GGUF builds, and whether 27B had become the practical sweet spot. By crawl time on April 25, 2026, the thread had 1,688 points and 603 comments.