DeepSeek-V4-Flash-0731 has 304 billion parameters
Controlling reasoning effort in LLMs is possible
OpenAI advances mathematics and computer science
Echo achieves Fable-level results at lower cost
Co-designing AI model attention improves inference
Smevals evaluates models, prompts, and harnesses
Stateless MCP specification updated
Datasette-agent 0.4a0 adds browser task mechanism
OpenAI reduces Codex model context size
Llm-mcp-client 0.1a0 released
AI coding agents modernize scientific computing
Building abundant intelligence with full-stack approach