Qwen3.8 Preview Hands-On Test
青稞AI · wechat · 2026-07-20
This post shares a hands-on experience with Qwen3.8 preview: the author integrated it into their AGIBar / mini-program project to develop complex system features like shared wallet and checkout.
Key Information
- The model is Qwen3.8 preview; the author estimates about 2.4T parameters and accessed it via Alibaba's TokenPlan.
- They provided an access method similar to ClaudeCode and a request address mapping approach, suggesting the model can be mapped directly to qwen3.8-max-preview.
- In real projects, the model was used to break down complex requirements and implement them, including wallet recharge, permissions, invoices, multi-person organizational deductions, WeChat/bank transfers, etc.
Hands-On Experience
- The author notes that the model completed quite complex system development tasks within about 1 hour.
- Self-measured generation speed roughly 84.5 words/second, which they equate to 84.5 tok/s.
- Conclusion: It can already help complete complex system development with sufficient speed, but still lags behind other top models in their view.
Related event: Qwen3.8-Max Preview Tested: Strong Coding but Slow Thinking(13 posts)→
More from coding & agent
- Gergely Orosz: Shipping 10x PRs With AI Agents, Sites Fill With Small Regressions — ducha_aiki · 2026-09-11
- Same Echo Maze prompt, three frontier models: all passed visually but shipped the same hidden bug — eyishazyer · 2026-09-11
- Astra storyboards plus Minimax H3 per-shot generation boost video success rates — Hailuo_AI · 2026-09-11
- Codex tip: use Sol with Astra and Luna sub-agents to save usage — pvncher · 2026-09-11
- agents-best-practices: a provider-neutral Agent Skill for designing and auditing agentic harnesses — tom_doerr · 2026-09-11
- Cognition's SWE-2 uses a KKT duality argument in RL to shift the effort Pareto curve — YouJiacheng · 2026-09-11