New - local AI
Kimi K3 now runs locally - Unsloth's 1-bit quant shrunk it 62%
The strongest open model to date (2.8T MoE) fits in RAM on a 748GB DGX Station at 594GB (-62% from 1.56TB), ~78.9% accuracy retained. On a Mac, it loads via SSD memory-mapping with a ~128GB working set - slower, but it runs.
Track the latest open-weight models, agent harnesses, and autonomous agents, then see which ones fit your rig. Honest speed estimates, cloud-pricing comparisons, and a source-cited tracker of who's running what - so no vendor or government order can switch off the model you depend on.
Already know your rig? Pick it here and see exactly which models you can run locally, with honest speed estimates and cloud-pricing comparisons.
Sign in with GitHub to save your hardware. New here? We'll guide you through picking your rig - Mac, multi-GPU (up to 8x), or custom specs - then show you exactly which models you can run locally.
Sign in with GitHub - freeThe Information: engineers evaluating Moonshot AI's Kimi K3 (2.8T open-weight, released 2026-07-16, $3/$15 per MTok) for Copilot features currently on GPT/Claude, citing strong coding benchmarks and ~60% lower inference cost. Not officially confirmed; evaluating, not deployed.
Bloomberg: Microsoft replacing OpenAI/Anthropic models with in-house MAI models in Excel and Outlook to cut inference spend; a tuned MAI variant claims GPT-5.4 parity at up to 10x efficiency. Proprietary (not open-weight), but the same frontier-API cost pressure.
Self-hosted Llama 3.1 8B on Kubernetes automates support-ticket creation and resolution drafts for the ad-tech platform; 80% less time to create tickets.
Mistral Medium 3.5 (128B) for up to 100k French public-sector agents under a 4-year, EUR 140M framework; on-prem SecNumCloud option for sovereignty.
Self-hosted Codestral in its RAISE/SovBox coding assistant for regulated aerospace, defense and public-sector clients; code-completion accuracy 50% -> 90%.
Statuses: confirmed official source reported credible third-party, not officially confirmed testing evaluating, not deployed.
Curated and source-cited, not a scraper. Built a rig worth sharing? Browse shared builds →.
