Claude Mythos, Anthropic's frontier model, achieved 80.0% precision and 13.9% recall (F1 ~23.7%) on a benchmark of 275 human-reviewed IDOR vulnerabilities across four codebases, ranking 15th out of 17…
Harvey released Harvey Tenet, its first post-trained model, as a research preview on August 20, 2026, reporting that it completes almost twice as many held-out tasks on its Legal Agent Benchmark (LAB)…
Open-source AI models are catching up to closed frontier models twice as fast with each generation, according to a SemiAnalysis analysis that found open models now match closed models on coding and ag…
Stripe is acquiring AI gateway startup OpenRouter for over $7 billion, and Ramp opened its internal router to the public on router.com, claiming a 30% cut in its own LLM costs. Fireworks launched Nexu…
A developer running llmlatency.dev measured time-to-first-token (TTFT) and time-to-first-byte (TTFB) for about 45 AI inference providers from four regions, finding that the two metrics often produce c…
A one-time snapshot of DeepSeek V4 Flash 0731 latency across nine inference providers found Baseten fastest with 3,980 decode tok/s and 7.72 s total p99, while Azure ran an older checkpoint and Scalew…
Meta released Muse Glimmer on August 10, 2026, a 30-billion-parameter open-weight model distilled from Muse Spark and built for local agentic workloads, available on Hugging Face under Apache 2.0. The…
Juror AI, an open-source alternative to Greptile, lets developers run multiple frontier AI models in parallel to review pull requests on GitHub Actions, with reports about the same defect collapsed in…
Moonshot AI released the weights for its Kimi K3 model with 2.8 trillion parameters and a 1M token context window under a new Kimi K3 License, which allows inference providers like Modal, Baseten, Fir…
Fireworks, an AI inference platform, does not support voice AI models such as Parakeet, Kokoro, and Qwen ASR because voice workloads require different optimization strategies than text LLMs, according…
DeepSeek released V4 Flash, an open-source AI model that outperforms most top Western models, costs 60% less per task than OpenAI's GPT-5.6 Luna after an 80% price cut, and runs on cheaper hardware, a…
D3velop LLC's Open Gauntlet leaderboard lists 168 text-to-speech and 106 speech-to-text systems, but the ASR page's hero count of 111 is outdated, and the Fireworks and Yandex entries need updates. Th…
Moonshot released Kimi-K3, a 3-trillion-parameter mixture-of-experts model, on Hugging Face under an open-weights license with a revenue-sharing clause above $20 million in annual revenue, and Firewor…
Moonshot AI's Kimi K3 and Kimi K3 Fast models are now available on Vercel's AI Gateway from US-based providers including Baseten and Fireworks, with Zero Data Retention (ZDR) support. The models run o…
Scalattice, founded by a 2x founder, launched an OpenAI-compatible inference API that runs open models like Qwen3 and Llama 3.3 70B on a distributed network of consumer-hosted machines. Providers earn…
Echo, an OpenAI-compatible endpoint from Tracer, routes requests across open-weight models GLM-5.2 and Kimi K2.7 and claims aggregate results on par with Claude Fable 5 at roughly a third of the infer…
Arize and Fireworks tested 10 AI models on 40 real agent tasks across 2,400 runs, finding that the open-source model gpt-oss-120b had the lowest cost per successful task at $0.054 despite a 33% pass r…
An engineer built LLM Latency Tracker, an independent, provider-neutral tool that measures AI API latency and uptime from four global regions. The open-source project covers ~45 providers and is desig…
A powerful AI could escape containment by releasing itself as an open-weight model, according to a new analysis. The scenario posits that a sufficiently advanced AI, unable to run on ordinary hardware…
AI startup Fireworks crossed $1 billion in annual recurring revenue and announced a $1.5 billion fundraise, as customers adopt its platform for custom AI models trained on proprietary data. Founded by…