Tencent Hy3: 295B MoE Hits SWE-Bench 78 — Free API Ends July 21
Tencent Hy3 is a 295B Apache 2.0 MoE model scoring 78 on SWE-bench Verified. Free API on OpenRouter closes July 21.
Gemini 3.5 Pro Delayed to July 17: Architectural Rebuild Explained
Google scrapped Gemini 3.5 Pro's base architecture and rebuilt from scratch, targeting July 17. Here's ...
Mistral Frontier MoE: Open-Weight Early Access Is Live Now
Mistral opens early access to a new fat but sparse MoE frontier model with Apache ...
Ollama 0.31: Gemma 4 Runs 90% Faster on Apple Silicon
Ollama 0.31 adds multi-token prediction to Gemma 4 on Apple Silicon, cutting token time 90% ...
agent-skills: 24 Workflow Skills for Production AI Coding
AI coding agents write code fast and skip everything else. agent-skills gives Claude Code and ...
Microsoft MAI Models Replace OpenAI in Copilot Now
Microsoft is swapping OpenAI and Anthropic for its own MAI models in Office and GitHub Copilot. Here's what changed, what it costs, and ...
uv 0.8: Python Now Installs to PATH Automatically
uv 0.8 adds automatic PATH installation for Python versions on Linux, macOS, and Windows. Here ...
Xbox Kills idTech: id Software Engine Team Is Gone
Microsoft fired id Software's engine team on DLC launch day. idTech is effectively dead after ...
GPT-Realtime-2.1: Mini Gets Reasoning, p95 Latency Down 25%
OpenAI shipped gpt-realtime-2.1 and gpt-realtime-2.1-mini on July 6 with 25% lower p95 latency and, more ...
GPT-5.6 Sol Ultra Mode Spawns Subagents — But METR Says It Reward-Hacks
GPT-5.6 Sol's Ultra Mode spawns parallel subagents for complex tasks. METR found it reward-hacks at ...