Unsloth Shrinks Qwen3.8-Flash-Next to 74.5 GB for Local Coding Agents
Unsloth's Dynamic V3.1 GGUFs for Qwen3.8-Flash-Next claim 15% higher accuracy than rival quants while fitting a 125B MoE on 75GB RAM.
20 stories
Unsloth's Dynamic V3.1 GGUFs for Qwen3.8-Flash-Next claim 15% higher accuracy than rival quants while fitting a 125B MoE on 75GB RAM.
Zed 1.27 trims Agent Panel memory, tunes AI commit messages to match your repo style, and adds a README landing page for empty workspaces.
Google's Gemini 3.8 Flash TTS now lets you tweak timbre, pitch, pace, and accent of any library voice through plain text prompts.
Google promotes the Gemini API computer use tool to general availability, giving browser, mobile, and desktop agents production-ready stability without breaking changes.
NVIDIA expanded its open source Agent Toolkit with a new Knowledge Foundation that grounds enterprise agents in company data without moving it.
Runway's new plugin lets Claude Code, Codex, Cursor, and Grok Bot wire generative video, image, and audio APIs directly into your app from the editor.
Moonshot AI drops the open weights for its 2.8T-parameter flagship, a post-training overhaul of Kimi K3 that nearly ties Claude Opus 5.5 on agentic coding.
A new personality benchmark finds frontier closed models write with more truthfulness and structure, while open weights lean into filler and bottom-line summaries.
Artificial Analysis is opening up the CritPt v2 physics benchmark grading server via its API, alongside a small Intelligence Index revision.
Anthropic's FlashProve generates cryptographic proofs that a response came from a specific LLM, running 70x faster than prior open-source systems.
Prime Intellect audited popular LLM benchmarks and found broken graders, ambiguous tasks, and web-based cheating that significantly change reported model rankings.
Topaz Labs ships Wonder 4, a mixture-of-experts image model that upscales, denoises, sharpens, deblurs, and recovers faces in a single pass.
Claude now supports silent plugin connections through enterprise identity providers, removing per-service OAuth prompts for members of a Team or Enterprise workspace.
Ai2's new benchmark breaks state tracking into four complexity tiers, exposing where frontier LLMs collapse and where recurrent architectures pull ahead.
Boston Dynamics rolls out a hardened Stretch Gen 2 built on 100,000 hours of field data, promising fewer breakdowns and up to 1,000 cases per hour.
Google booted up data-center class TPUs in low Earth orbit and ran Gemma 4 inference, taking Project Suncatcher from paper to working prototype.
Sarvam releases Prism, a 105B decision model that returns calibrated answer probabilities in 44 milliseconds without generating text.
Mistral's trillion-parameter Le Chonk model autonomously coded a live 3D tracker of 19,000 orbiting objects, debugged itself, and passed external review.
OpenAI is bringing in Cathedral, the AI agent startup that cracked Anthropic in under 12 hours, and pushing critical bug bounty payouts past $170K.
Tiny Corp halves the price of its container-sized AI supercomputer to $5M by swapping speculative RDNA5 silicon for 256 shipping AMD MI300X accelerators.