NVIDIA, MIT, and NTU's SoL-Pi has a coding agent generate 152 efficiency research proposals and keep four, cutting tokens nearly in half on the authors' EdgeBench suite.
The world's dominant AI chip maker is acquiring the platform where 18 million developers find, share, and deploy open-source models. The deal closes in early 2027 — pending regulatory review.
Z.ai's new open-weight model hit 84.5% on CyberGym and 54.4% on ExploitBench, and has already identified a serious security flaw in the Cursor code editor.
Alibaba's biggest model yet went from preview to general availability on August 3. Qwen3.8-Max is a 2.4-trillion-parameter mixture-of-experts model with a 1M-token context window and strong agent benchmarks — though independent evaluations haven't caught up yet.
Responding to accusations of opposing open-source AI, Anthropic's CEO clarified the company's actual position: open-weight models are a public good, with one serious exception.
Moonshot AI released the full open weights for Kimi K3 on July 26-27. At 1.4TB in MXFP4 format, it's the largest open-weight model ever published. Together AI and Modal launched day-0 hosted access for teams without Blackwell hardware.
Thinking Machines Lab released Inkling, a 975B-parameter open-weight model trained on 45 trillion tokens across text, image, audio, and video. Former OpenAI CTO Mira Murati is betting enterprises want AI they can customize, not just rent.
Kimi K3 has 2.8 trillion total parameters, a 1M-token context window, and native vision. It launched July 16 on the Kimi app and API. Open weights arrive by July 27.
Mistral released Leanstral 1.5 on July 2 — a 119B open-weights model built for Lean 4 formal verification that found five previously unknown bugs across 57 real open-source repositories.
DeepSeek V4 brings two model tiers, 1M-token context by default, and a hard cutoff on July 24 for legacy API names that millions of developers still use.
Cursor quietly acqui-hired Continue, the open-source GitHub Copilot alternative, in June. The product is shutting down: billing is off, the repo is read-only, and cloud data gets permanently deleted on July 15.
Google shut off Gemini CLI for individual users on June 18, 2026. The Apache 2.0 open-source tool that launched in 2025 is replaced by Antigravity CLI, a closed-source Go-based agent terminal. Enterprise customers keep their access.
Cohere released North Mini Code on June 9, 2026 under Apache 2.0. The 30B/3B mixture-of-experts model targets enterprise teams who want a capable agentic coding model they can run on-premises without vendor dependency.
Zhipu AI released GLM-5.2 on June 13, 2026 with a usable 1-million-token context window, two thinking-effort levels, and a promise of MIT open weights the following week. The release came two days after the US government ordered Anthropic to cut foreign access to its Fable 5 models.
Moonshot AI released Kimi K2.7-Code on June 12: a trillion-parameter open-source coding model that cuts reasoning token usage by 30% compared to K2.6. Every benchmark number comes from Moonshot's own proprietary evals.
Google open-sourced the Colab CLI on June 5, letting developers and AI agents provision A100s and H100s, run Python scripts, and download results without touching a browser. A bundled COLAB_SKILL.md makes it ready for Claude Code, Codex, Gemini CLI, and other terminal agents out of the box.
JetBrains released Mellum2 under Apache 2.0 on June 1. It's a 12B parameter mixture-of-experts model with only 2.5B active per token, designed for the repetitive, latency-sensitive tasks inside AI coding pipelines: routing, RAG, sub-agents, and tool use.
On May 19, Google announced that Gemini CLI's free access ends June 18. The replacement is Antigravity CLI, a closed-source Go tool with a near-useless 20-request/day free tier. Community reaction to the open-source bait-and-switch has been sharp.
GitHub published the full source code for its Eclipse IDE plugin on May 21, 2026. The repository at github.com/microsoft/copilot-for-eclipse contains roughly 15,000 lines of Java covering code completion, Next Edit Suggestions, chat, and agent mode.
Moonshot AI released Kimi K2.6 on April 20. The open-weight model scores 80.2% on SWE-Bench Verified and 66.7% on Terminal-Bench 2.0, with pricing of $0.60 per million input tokens on the official API. It's a meaningful upgrade over K2.5 on every benchmark that matters for coding agents.
Xiaomi released MiMo-V2.5-Pro on April 27: an open-source 1-trillion-parameter MoE model that scores comparably to Claude Opus 4.6 on coding benchmarks while using 40–60% fewer tokens per task. Weights are freely available on Hugging Face.
The open-source VS Code coding agent is ending its extension, Cloud, and Router services on May 15, 2026. The team says the IDE era is over and is betting on Roomote, a cloud-based autonomous agent that works through Slack, GitHub, and Linear.
OpenAI updated its open-source Agents SDK on April 15 with native sandbox support, a new model-native harness for file and shell operations, configurable memory, and workspace portability via S3, GCS, Azure, and Cloudflare R2. Python ships now; TypeScript is coming.
A leaked model ID revealed that Cursor's Composer 2 is fine-tuned from Moonshot AI's Kimi K2.5. Here's the full timeline of the controversy, the license questions, and how it resolved in under 24 hours.
ESC
Start typing to search across tools, news articles, and reviews
No results found
Optional analytics and social embeds help us improve the site. It works fully without them.
Privacy
Privacy Settings
Choose which optional features you want to enable.
Your browser sent a global privacy opt-out signal. We respect that by default.
