Uploaded August 2026 | Updated September 2026, 2 weeks ago
Qwen 3.8 Max: 2.4T Params, Agentic Long-Horizon Tasks, Benchmarks + Web App Demos
I cover Qwen’s new Qwen 3.8 Max release, a 2.4T-parameter multimodal model with 95B active params that’s expected to go open-weight soon, and I compare excitement around the upcoming 27B model that can run on consumer hardware. I review benchmarks (including a top-five Arena spot and coding/agentic performance comparisons), then dig into agentic long-horizon capabilities like creating and evolving its own harness (“Oh My CLI”) and examples like autonomous chip design. I run my own tests: an ISS tracker web app, a Pokémon encyclopedia UI, a crowd animation task where it “cheats” by overlaying text, and a Los Angeles tourist 3D map. I also discuss first-party API pricing vs Kimi K3, plus the open-weight licensing uncertainty and enterprise vs general-public tradeoffs.
Announcement: https://x.com/Alibaba_Qwen/status/2084100707423289643
oh-my-cli: github.com/qwen-code-dev-bot/oh-my-cli
Blogpost: qwen.ai/blog?id=qwen3.8
My voice to text App: whryte.com
Website: engineerprompt.ai
RAG Beyond Basics Course:
prompt-s-site.thinkific.com/courses/rag
Signup for Newsletter, localgpt:
https://tally.so/r/3y9bb0
00:00 Qwen 3.8 Max
01:12 Benchmarks and Rankings
01:54 Agentic Long Horizon Claims
03:12 Oh My CLI Harness
04:01 Website Self Demo
04:56 ISS Tracker Web App Test
06:50 Pokemon Encyclopedia Demo
07:16 Crowd Animation Cheat
08:37 LA Tourist 3D Map
09:24 API Pricing vs Kimi
09:59 Open Models and Licensing
Qwen 3.8 Max: 2.4T Params, Agentic Long-Horizon Tasks, Benchmarks + Web App Demos
I cover Qwen’s new Qwen 3.8 Max release, a 2.4T-parameter multimodal model with 95B active params that’s expected to go open-weight soon, and I compare excitement around the upcoming 27B model that can run on consumer hardware. I review benchmarks (including a top-five Arena spot and coding/agentic performance comparisons), then dig into agentic long-horizon capabilities like creating and evolving its own harness (“Oh My CLI”) and examples like autonomous chip design. I run my own tests: an ISS tracker web app, a Pokémon encyclopedia UI, a crowd animation task where it “cheats” by overlaying text, and a Los Angeles tourist 3D map. I also discuss first-party API pricing vs Kimi K3, plus the open-weight licensing uncertainty and enterprise vs general-public tradeoffs.
Announcement: https://x.com/Alibaba_Qwen/status/2084100707423289643
oh-my-cli: github.com/qwen-code-dev-bot/oh-my-cli
Blogpost: qwen.ai/blog?id=qwen3.8
My voice to text App: whryte.com
Website: engineerprompt.ai
RAG Beyond Basics Course:
prompt-s-site.thinkific.com/courses/rag
Signup for Newsletter, localgpt:
https://tally.so/r/3y9bb0
00:00 Qwen 3.8 Max
01:12 Benchmarks and Rankings
01:54 Agentic Long Horizon Claims
03:12 Oh My CLI Harness
04:01 Website Self Demo
04:56 ISS Tracker Web App Test
06:50 Pokemon Encyclopedia Demo
07:16 Crowd Animation Cheat
08:37 LA Tourist 3D Map
09:24 API Pricing vs Kimi
09:59 Open Models and Licensing
