Improved performance and model support with GGUF
Ollama 0.30 adds GGUF model compatibility through llama.cpp, up to 20% faster performance on NVIDIA hardware, and Vulkan enabled by default for broader GPU support on AMD and Intel devices.
Releases, research, and ideas for developers and product teams. Updated every hour.
Ollama 0.30 adds GGUF model compatibility through llama.cpp, up to 20% faster performance on NVIDIA hardware, and Vulkan enabled by default for broader GPU support on AMD and Intel devices.
Ethan Mollick announced Co-Existence, a new book due October 20, arguing that AI has moved from cooperative chatbots to partly autonomous agents that now outperform humans in some tasks, and that Anthropic reports AI writes 80% of its code.
NVIDIA Nemotron 3 Ultra, a 550 billion parameter open model with 55B active, is now available on Ollama's cloud for long-running agentic workflows. It offers 1M token context and is optimized for NVIDIA's NVFP4 format, with Ollama citing up to 30% cost savings versus other leading open models.
Import AI 459 covers a paper estimating the US AI economy's quality-adjusted output grew roughly 2,290 percent in 2024 and 2,271 percent in 2025, UK AISI research on why automated alignment is difficult, Stanford-led release of the 100M-image GPIC dataset, and Biohub's ESMFold2 protein model.
Mistral unveiled a new AI stack for industrial engineering at the AI Now Summit 2026, with Airbus, BMW Group, and ASML as partners. It also updated its agentic tool Vibe for long-horizon tasks and announced a 10 MW data center in Les Ulis, France, scheduled to open in Q3 2026.
Mistral AI has launched Vibe, a unified AI agent for work and coding that replaces Le Chat, with Work Mode and Code Mode plus a new VS Code extension and CLI updates. It is available at chat.mistral.ai on Free, Pro, Team and Enterprise plans.
Mistral AI has released Search Toolkit in public preview, an open source composable framework for building production search pipelines for AI applications that unifies ingestion, retrieval, and evaluation in one shared interface.
OpenJarvis v1.0, an open-source framework for building personal AI agents that run on local hardware, is now available with built-in support for Ollama. It is built by Stanford's Hazy Research and Scaling Intelligence labs as part of their Intelligence Per Watt research into efficient local AI.
Mistral announced physics AI, a new class of models that predict physical system behavior from geometry and boundary conditions, integrating Emmi AI into its enterprise platform for industrial engineering. Partners named include ASML, Airbus, Safran, and Siemens Energy.
Mistral AI published a blog post outlining physics AI research areas it is pursuing following its acquisition of Emmi AI, listing several papers and models covering transonic aerodynamics, computational fluid dynamics, automotive and aerospace applications, plasma turbulence, and industrial simulation. Current scientific results from these efforts are sparse and remain at a preliminary stage.
Elon Musk lost a $150 billion suit against OpenAI and Sam Altman, and Google updated its Gemini app at IO 2026 in a move described as taking on ChatGPT and Claude.
Two education studies with overlapping research teams, including Wharton researchers, found that students given plain ChatGPT did worse on exams without it, while students given an AI tutor that personalized problem sequences scored 0.15 standard deviations higher on a final exam taken without AI help.
Import AI issue 458 consists of a long essay based on a 2026 Cosmos HAI Lab Lecture given at the University of Oxford, arguing that continued AI progress forces a choice between exploring the future or retreating from the present, plus a fictional story about a positive singularity.
Episode 246 of the Last Week in AI podcast, recorded May 22, 2026, covers Google I/O announcements including Gemini 3.5, the Gemini Spark agent, and Gemini Omni, plus Elon Musk losing his OpenAI lawsuit on statute-of-limitations grounds and an OpenAI result on an 80-year-old Erdős geometry problem.
Mistral AI has entered into a definitive agreement to acquire Austrian Physics AI company Emmi AI. Emmi's co-founders and more than 30 researchers and engineers will join Mistral's Science and Applied AI teams in May.
Mistral AI released Mistral Medium 3.5 in public preview, a 128B open-weight model that powers remote coding agents in Mistral Vibe and a new Work mode in Le Chat. It is available on Pro, Team, and Enterprise plans, priced at $1.5 per million input tokens and $7.5 per million output tokens.
Mistral AI released Connectors in Studio, available in Public Preview, letting developers use built-in and custom MCP connectors via API/SDK across model and agent calls. The release adds direct tool calling and human-in-the-loop approval flows.
Last Week in AI episode 245, recorded May 13, 2026, covers OpenAI's new voice intelligence API features, a Thinking Machines conversational system preview, and Anthropic's Claude for Legal launch. The episode also discusses a Chinese grey market selling Claude API access and Isomorphic Labs raising $2.1 billion.
Import AI 457 covers SentinelOne's teardown of the fast16.sys virus that tampered with high-precision calculation software, Tilde Research's finding that the Muon optimizer can cause neuron death in MLP layers and its proposed Aurora optimizer, a position paper on positive alignment, and Prime Intellect tests of autonomous AI research agents on the nanoGPT speedrun.
A survey of open-weight LLM architecture changes from April to May shows a common push to cut long-context costs, including cross-layer KV cache sharing in Google's Gemma 4 E2B and E4B, which saves about 2.7 GB and 6 GB respectively at 128K context in bfloat16.