Mouse, an agent harness built on the OpenCode engine, scored 24/30 on the FrontierHarness benchmark, the highest pass rate among 13 harnesses tested with Kimi K3 on 30 coding tasks. Mouse passed 18/21…
GitHub will deprecate Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7 across all GitHub Copilot experiences on October 2nd, 2026, urging users to migrate to suggested alternati…
Conversion, a marketing intelligence company, announced it built a self-improving text-to-query agent that converts natural-language audience descriptions into editable filters, using a tool-using age…
Anthropic's Claude Fable 5.1 (Max), launched on September 1, 2026, topped the Code Arena: WebDev leaderboard with 1,765 points, outpacing Alibaba's Qwen3.8 Max by 77 points. The model, which builds on…
Open-weight AI models like DeepSeek V3.2 (685 billion parameters, ~685 GB) and Kimi K3 (2.8 trillion parameters, ~3 TB) are increasingly difficult to run locally due to hardware constraints, prompting…
Moonshot AI has confidentially filed for a $3 billion Hong Kong IPO at a $50 billion valuation, a move that reflects Beijing's strategy to keep frontier AI capital under domestic control by requiring …
Google launched Gemini 3.8 Flash, its third Flash model in six weeks, claiming flagship-level coding performance at lower cost, with CEO Sundar Pichai announcing the release. The model is priced at $0…
Anthropic's Claude Opus 5.1, released September 1st, scored 92.5% on an independent coding benchmark, the highest recorded on that test, surpassing GLM 5.3 (91.1%), Opus 5 (82.5%), and Qwen 3 Max (81.…
Inco AI launched its inference platform in public beta, claiming the fastest output speeds on Artificial Analysis leaderboards for four open models: Kimi K3, MiniMax M3, GLM 5.3, and GLM 5.3 Flash. Th…
A new evaluation method using typed ontologies caught 100% of citation defects in LLM-generated research memos that a standard LLM judge missed, according to an experiment by antithetical-labs. Removi…
A coalition of 25 companies including Nvidia, Microsoft, and Meta published an open letter on July 24 titled 'Open Weights and American AI Leadership,' opposing efforts by certain frontier labs to res…
China's Moonshot AI's open-weight model Kimi K3 uncovered 7,958 potential security issues across 501 Bitcoin-related open-source projects in about 108 hours, with 1,280 rated high or critical severity…
Baseten's technical breakdown of LLM inference efficiency identifies two categories of engineering choices: those that trade latency for throughput, such as batch sizing, tensor parallelism, expert pa…
Inference engineering techniques for large language models (LLMs) can be categorized into those that manage tradeoffs along an efficient frontier and those that push the entire frontier outward, accor…
Between July 27 and August 25, 2026, five Chinese labs released four frontier open-weight AI models—Kimi K3, DeepSeek V4-Flash-0731, Qwen3.8-2.4T-A95B, and GLM-5.3-Flash—each with distinct architectur…
DogLM, a new benchmark created by Mike Ushakov, tests whether large language models (LLMs) let players pet a dog in AI-generated video games, inspired by the 'Can You Pet the Dog?' rule. In tests of 1…
A 2.78-trillion-parameter Kimi K3 model ran on a single CPU with 8.24 GB of RAM using a 176 KB portable C99 engine, though it requires a 1.56 TB checkpoint on disk and runs at 26.5 seconds per token a…
GitHub's new global model policy, enforced since August 26, flips Copilot's default from opt-in to opt-out, enabling any generally available model that admins never explicitly configured, with gradual…
SemiAnalysis's ClusterMAX 3.0 testing found that most neoclouds have serious security vulnerabilities, with five frightening patterns identified. The company urges neocloud operators and users to upda…
Chinese AI models are closing the performance gap with US rivals while undercutting them on price, forcing Silicon Valley to cut prices, according to users and analysts. Zhipu AI's GLM-5.2 costs US$4.…