Frontier AIFrontier Flux Originals
Full stories written by Frontier Flux. Listed here are the originals from the last 30 days. The same pieces also appear on the homepage and the morning brief. Each story cites the primary post or outlet.
Stories older than 30 days leave this list for the archive of originals.
On 10 September 2026 OpenAI Developers said the Agents API is in public beta: a managed Codex harness for cloud agents, with OpenAI handling orchestration, long-running sessions, and context management. OpenAI says there is no extra Agents API fee beyond tokens and tools.
DeepSeek released DeepSeek-V4.1-Flash with MIT weights on Hugging Face and API id deepseek-flash. The lab describes a 552B MoE Causal Encoder-Decoder that activates 8B parameters on prefill and 16B on decode, plus a 890-byte global KV cache per token.
On 8 September 2026 OpenAI said ChatGPT Images 2.5 is rolling out to all ChatGPT, ChatGPT Work, and Codex users, with two API models: GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst. The lab reports up to 50% lower generation latency versus Images 2.0, plus Sketch, templates, and on-image comments in ChatGPT.
On 8 September 2026 OpenAI published a research note saying a group of agents, using an unnamed internal model it describes as significantly more capable than GPT-6 Astra, produced an analytical proof and a Lean formalization that three-dimensional incompressible Navier-Stokes can develop a singularity in finite time. The company says it does not intend to claim the Clay Millennium Prize.
On 6 September 2026 OpenAI chief scientist Jakub Pachocki published An Alien Mind. He writes that no lab has solved alignment and monitoring enough to keep scaling at maximum speed for much longer, and he hopes voluntary slowdowns become common until shared safety bars exist.
On 4 September 2026 OpenAI said GPT-6 Astra is now available to Pro, Enterprise, and Business Premium users in ChatGPT Work and Codex, and live in the API. Plus and remaining Business users might wait a few more days. The lab calls Astra its most capable broadly deployed model and its first Critical cybersecurity model under the Preparedness Framework.
On 3 September 2026 the Institute of Foundation Models released K2 Horizon, a six-model fleet from 0.9B to 375B-A23B, with Apache 2.0 Hub tags. IFM says the 0.9B, 3.7B, and 7B models set size-class state of the art. The 375B card says the final checkpoint is live while some data and training code still will be released.
On 2 September 2026 Google launched Gemini 3.8 Flash as its new workhorse at the same introductory $0.75/$3.75 per million tokens as 3.7 Flash, plus Gemini 3.8 Flash Cyber for trusted defenders through the new Fairwind Program.
On 1 September 2026 Meta Superintelligence Labs launched Muse Voice Transcribe, its first real-time audio perception model, with streaming ASR, 20-plus-speaker diarization, and endpointing. It is live on the Meta Model API, Meta AI for Mac, and Muse Code. Rankings Meta cites are as of that date.
On 31 August 2026 Google Research posted TimesFM-3, a 330 million parameter time-series model it says can forecast several related series in one forward pass. PyTorch weights are on Hugging Face under a non-commercial license. GitHub tagged v3.0.0 on 28 August.
Showing 10 of 18 stories
Free
Free daily email of model releases, benchmarks, and lab notes from the last 24 hours across OpenAI, Anthropic, DeepMind, Meta, NVIDIA, and more. Capability-first, not company drama.
Opens Buttondown to finish signup. Confirm by email if asked. No ads. Unsubscribe anytime.
