Everybody's saying slow down.
I think something else is going on 🤔 Closed labs don't publish. You never see inside their architecture. Open labs publish everything, and in the space of about a week Qwen and DeepSeek both shipped the same idea: move a huge share of the
GPT-6 Luna is LAUNCHING in September
Likely on par with GPT-5.6 Sol, but at a fraction of the cost a very strong competitor to DeepSeek models most likely launching alongside Sol and Terra I am incredibly excited
There is an alternative to Claude Code on GitHub, and the warehouse is called free-claude-code.
The interface is still the same programming assistant. The person working down there is not a genuine Claude. Replace with alternative models such as DeepSeek, Kimi, and Gemini. Use the free quota of each company. Stop when the quota is gone. It’s not an official gift of Claude. Address: The monthly fee seems expensive, so I’ll take this one for a while.
Qwen3.8 flash next is my new favorite local model
One-shot voxel turtle, zero retry. ran it through the deepseek harness (qwen + deepseek combo goes crazy) tried the same thing on other harnesses and it didn't click. no idea why all local on my spark at 40-50 tok/s. open
You have reached the end of the archive
All of deepseek