Inkling: Thinking Machines' Open Multimodal AI
Thinking Machines released Inkling, a 975B-parameter open-weights Mixture-of-Experts model that reads text, images, and audio. You can self-host it or fine-tune it via Tinker.
Kimi K3: World's First Open 3T-Class AI Model
Moonshot AI has officially launched Kimi K3, the world's first open 3T-class model: a 2.8T-parameter MoE with a 1M-token context window and open weights arriving July 27.
Gemini 3.5 Pro Reportedly Targets a July 17 Launch
Google's Gemini 3.5 Pro is reportedly targeting a July 17 general-availability launch, according to leaked launch plans and third-party reporting rather than any official Google post.
Run Qwen3.6 2.5x Faster Locally With Unsloth NVFP4
Unsloth released NVFP4 quantized versions of Qwen3.6 that run up to 2.5x faster, with the 27B model fitting on a single 24GB GPU.
Run GLM-5.2 (744B) Locally on 25GB RAM: Colibri
A developer released Colibri, a pure-C engine that runs GLM-5.2 (744B MoE) on a 25GB-RAM machine with no GPU by streaming experts from an NVMe SSD.
Subagentmaxxing: Cheap AI Coders as Claude Subagents
Subagentmaxxing is a new open-source CLI that runs Codex, Cursor, and Grok as subagents under Claude Code, keeping flagship planning while cutting execution costs up to 25 times.
Ollama Hits 9M Devs: Your Local Open-Model Workflow
Ollama raised $88M and now serves 8.9M developers. Here is how to run open AI models locally, when to reach for Ollama Cloud, and a full getting-started workflow.
Mistral Studio Adds Version Control for Prompts
Mistral added version control for prompts and skills to Mistral Studio on July 9, 2026, with immutable versions, rollback, ownership tracking, and audit logs.
Run Any Hugging Face Model at vLLM Speed
Hugging Face and vLLM shipped a native-speed transformers backend on July 8, 2026, serving almost any Hugging Face model at full vLLM speed with a single flag.
OpenAI GPT-Live: Full-Duplex ChatGPT Voice
OpenAI launched GPT-Live on July 8, 2026, a pair of full-duplex voice models that listen and speak at the same time, replacing Advanced Voice Mode for every ChatGPT user.
Tencent Hy3: Open 295B Agentic Coding Model
Tencent open-sourced Hy3, a 295B mixture-of-experts model (21B active, 256K context) under Apache 2.0. It beats GLM-5.2 at roughly half the size and is free on OpenRouter until July 21.
LongCat-2.0: China's 1.6T Open-Weights Coding Model
Meituan open-sourced LongCat-2.0, a 1.6-trillion-parameter agentic coding model trained entirely on Chinese chips. The weights just landed on Hugging Face, and it beats GPT-5.5 on SWE-bench Pro.
Claude Fable 5 Is Now Available: A Builder's Guide
Claude Fable 5 is generally available again as of July 1, 2026: Anthropic's most capable widely released model for reasoning and long-horizon agents. What it is, the pricing, and the refusal-and-fallback change every builder must handle.
Claude Sonnet 5: Near-Flagship Power at a Lower Price
Anthropic launched Claude Sonnet 5 on June 30, 2026 as the new default model for Free and Pro, positioned close to Opus 4.8 at a lower price. Here is what changed, the tokenizer catch that resets your cost math, and how to switch your workflow.
Claude Tag: Anthropic's Always-On AI Teammate in Slack
Anthropic launched Claude Tag, an always-on AI teammate that lives in Slack with persistent memory, ambient mode, and autonomous async tasks.
Weave Router: One Proxy to Cut AI Coding Costs 40-70%
Workweave open-sourced Weave Router, a drop-in proxy that routes each prompt to the best model for Claude Code, Codex, and Cursor, claiming 40 to 70 percent lower costs.
DAI Studio Brings Visual Context Engineering to LLMs
Context engineering just got a dedicated workbench. On June 23, 2026, DAI Studio launched a free visual tool that lets you design, see, and version the exact context you feed into a large language model before it runs.
Sakana Fugu: One API to Orchestrate Top AI Models
Sakana Fugu is a multi-agent system that behaves like one model: send a request to a single OpenAI-compatible API and it routes across frontier models for you. Here is what it costs, how it scores, and why orchestration matters for builders.
Claude Fable 5 vs Opus 4.8: What Creators Get
Anthropic released Claude Fable 5, its most capable model, on June 9, 2026, free on paid plans through June 22. How it compares to Opus 4.8.
Xiaomi MiMo Hits 1000 Tokens Per Second on 1T Open Model
Xiaomi MiMo-V2.5-Pro-UltraSpeed claims 1000 tokens per second decode on a 1T MoE. Open-weights FP4 checkpoint plus a 2-week free API trial.
ICML 2026: AI Models Could Run on 97% Less Memory
New ICML 2026 research shows transformer models can share attention projections, achieving up to 96.9% KV cache reduction with minimal accuracy loss.
Nemotron 3 Ultra: NVIDIA's 550B Open-Weights MoE
NVIDIA released Nemotron 3 Ultra on June 1 2026: a 550B mixture-of-experts model with 55B active parameters, open weights on Hugging Face, with 5x faster inference and 30% lower cost than Nemotron 2.
Tesla V100 in a Gaming PC: 32GB Local LLM for £200
Developer Oscar Molnar installed a secondhand Tesla V100 SXM2 into his gaming PC alongside an RTX 4080, building a 32GB dual-GPU setup for under £200 total.
George Hotz: AI Agents Are a Slot Machine for Creative Work
George Hotz argues AI agents frontload impressive progress but stall on polish, creating a golden era of AI-generated slop that creators need to plan for.