The latest Model releases stories across our sources, prepared from the publishers’ own reporting.
Z.ai's GLM 5.3, a 753B-parameter mixture-of-experts model optimized for coding and long-horizon agentic tasks, is now available on Amazon Bedrock for eligible enterprise customers. Z.ai reported a CyberGym benchmark score of 84.5 at release.
Reflection AI unveiled Beam, its first open-weight frontier AI model, claiming it matches leading Chinese open models on reasoning benchmarks at lower compute cost. The Brooklyn-based startup says it will release Beam's weights and technical details this month.
Reka AI released a research preview of Rho-1, a 19-billion-parameter omni-model that processes and generates text, images, video, and robot control actions in one neural network. It trained on 320 H100 GPUs for about three months.
Aleph Alpha has released Kolibri, an open-weight German-English mixture-of-experts model with 78 billion parameters, about three billion active per token, trained on 768 B200 GPUs in Germany and Finland and published under the Apache 2.0 license.
Cloudflare released Clef and Clef-flash, decision models for AI agents that return probabilities for predefined answer options instead of generating text. Cloudflare reports median latency of about 39 milliseconds for Clef-flash and about 209 milliseconds for Clef, compared with just over 524 milliseconds for TypeSafe AI's Jev.
OpenAI published a guide on choosing GPT-6 family models, tuning reasoning effort, improving prompts and skills, coordinating tools, and preparing workflows for production. The guide is aimed at startups.
Ai2 has open-sourced AstaBrief 8B, a report-generation model built on Qwen3-8B that turns a research question and retrieved literature into a cited report. It is used as Fast mode in the Asta platform and released with its training data and example workflow.
Google announced Gemini 4 Argon, a frontier model with a 1-million-token output limit that is rolling out first to trusted cyber defenders through the Fairwind Program. The September 2026 roundup also includes Gemini 3.8 Flash, new voice and music models, Connected Apps, and science projects.
Black Forest Labs has released Flux 3 Image, the image component of its Flux 3 model family, supporting multi-step edits that leave the rest of an image unchanged. A free demo is available, API access is 50 percent off through October 8, and an open-weight version is expected in the coming weeks.
GPT-6 Astra Ultrafast is now available in the OpenAI API and to eligible ChatGPT Work and Codex users, running on NVIDIA Blackwell GPUs with up to 8x faster token generation than Astra Standard mode.
Ideogram has announced Ideogram 4.5, a model it says edits only selected parts of an image while keeping the rest unchanged. It offers four quality tiers from 0.8 to 22 cents per image at native 2K resolution, and an open-weight release is planned.
Amazon Web Services' Strands Labs released Strands Decider 2B, an open-source decision model inspired by TypeSafe's Jev. It sorts between pre-decided options and reports confidence, is small enough to run locally, and briefly topped the Jevbench ranking for its size.
Cloudflare released Clef and Clef-flash, two open-source decision models hosted on Workers AI, and debuted a reinforcement learning service for fine-tuning Clef. Clef leads the Jev Decision Index and the models are Apache 2.0 licensed on Hugging Face.
At OpenAI DevDay 2026, OpenAI launched Dots, always-on personal agents built on a never-ending GPT-6 Astra chat, alongside GPT-6.1 Sol, a Pro 500 plan, Sign in with ChatGPT, Plugins, ChatGPT Space and an Ultrafast mode. Google released benchmark scores for Gemini 4 Argon, which is not yet available to use.
TypeSafe AI, founded by former OpenAI researcher Diogo Almeida, released Jev, a decision-only model that returns typed probabilistic decisions instead of text. It evaluates typed questions in one parallel pass, returning Choice, Score and Noul answers with probability and confidence values; Vercel and Netlify integrated it quickly.
NVIDIA detailed a NeMo framework workflow for fine-tuning Nemotron 3.5 ASR on Saudi Najdi and Hijazi dialects using 133.7 hours of speech, reporting word error rate dropping from 55.05% to 29.96% on the target test split. NVIDIA also released Nemotron 3 Diarization for speaker-attributed transcription of up to eight speakers.
Google has released Gemini 4 Argon, which it calls its most powerful model yet, initially rolling it out to select cyber partners through its Fairwind Program. Google says the model was trained for defensive cyber work and claims it outperforms models from OpenAI and Anthropic on benchmarks.
Google unveiled Gemini 4 Argon, its first frontier model in over seven months, which matches OpenAI's GPT-6 Astra on an independent intelligence index but trails Anthropic's Claude Opus 5.5. Introductory API pricing is $2 per million input tokens and $10 per million output tokens.
Google announced Gemini 4 Argon, a new AI frontier model it says delivers frontier performance in complex workflows including software engineering, enterprise knowledge work, and cybersecurity defense. Google is initially limiting access to a set of trusted cyber defenders while engaging in a US government pre-release access process.
Google announced Gemini 4 Argon, claiming industry-leading performance in coding, knowledge work, and cybersecurity, but the model is not yet publicly available. Google says internal engineers are already using it and announced API pricing, with a phased release starting with trusted testers.