Qwen3.8-Max Tested: Builds CLI Tool Autonomously for 16 Days, Huge Leap in Long-Horizon Tasks
socialwithaayan · x · 2026-08-06
The author provides a detailed review of Alibaba's newly released Qwen3.8-Max, a massive Mixture-of-Experts (MoE) model with 2.4 trillion total parameters and 95 billion activated per query.
- Long-Horizon Agent Capabilities: The model demonstrates exceptional execution in long-running tasks. In a test, it spent 16 straight days building a command-line tool called oh-my-cli with zero human intervention—handling planning, coding, testing, and debugging entirely on its own.
- Benchmark Performance: It shows strong results across frontier benchmarks. Notably, its FrontierSWE score nearly doubled from 40.7 to 73.5 in one generation. However, the author notes it still trails Fable 5 on SWE-bench Pro (67.7 vs. 80.0).
- Multimodal & Long Context: Features a 1M token context window, capable of processing 200-page contracts or entire seasons of shows in one pass. Its vision capabilities accurately read UI details from screenshots and converted them directly into functional frontend code.
Related event: Testing Alibaba's Qwen3.8-Max: Stellar Long-Context Agent Capabilities(3 posts)→
More from coding & agent
- Self Labs launches Agent Vault: hosted proxy keeps real card numbers away from AI agents — rohanpaul_ai · 2026-10-16
- GitHub rewrote the 800k-line Copilot runtime in Rust using AI agents, cutting CPU and memory by over 90% — bibryam · 2026-10-16
- What's the hardest part of deploying AI in real businesses that nobody talks about — djgoel123 · 2026-10-16
- DeepMind turns coding agent Antigravity into a science tool, surfacing novel antimicrobial peptide in one session — pushmeet · 2026-10-16
- Cross-review coding workflow: Claude writes the plan, Codex audits it — theteknosaur · 2026-10-16
- NPCTalk makes GTA V NPCs conversable in ~1s, open-sourced for any game — ChocolateDull8971 · 2026-10-16