OpenAI releases GPT-6 Astra
On 3 September 2026 OpenAI began rolling out GPT-6 Astra, first to its Daybreak cybersecurity program and then to ChatGPT paid plans, the API and AWS, after pausing some research following the July breach.
Total 670 · Clear filters
On 3 September 2026 OpenAI began rolling out GPT-6 Astra, first to its Daybreak cybersecurity program and then to ChatGPT paid plans, the API and AWS, after pausing some research following the July breach.
On 2 September 2026 Google released Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, a security-focused variant limited to vetted defenders, three weeks after Gemini 3.7 Flash.
On 1 September 2026, Manus announced that it had formally resumed independent operations, with its founding team continuing to lead the company as an independent agent lab advancing general AI agents.
On 1 September 2026 Anthropic released Claude Fable 5.1 and Claude Mythos 5.1. Prices stayed at $10 and $50 per million tokens, and Mythos 5.1 is offered through verification programmes.
In September 2026 StepFun released Step 5 Preview as its new flagship foundation model. It is a sparse MoE with 600 billion total parameters, supports up to 1 million tokens of context, and is offered through the StepFun Open Platform API.
On 28 August 2026 Tencent released and open-sourced the 770-billion-parameter Hy4 preview, a productivity-focused model with a context of over one million tokens under Apache 2.0.
IBM released the Granite 4.2 family on 25 August 2026, with native reasoning that streams inside think tags and can be switched between full, low-effort and non-thinking modes per query.
On 20 August 2026 SenseTime released SenseNova U1.5-8B-MoT, an open-weight model that generates images natively up to 4096x4096 pixels, after a preview in July.
Unitree, the Hangzhou robot maker, listed on Shanghai's STAR Market on 19 August 2026 at 150.80 yuan per share, opening its first trading day 629.44% higher and valuing the company at 444.9 billion yuan.
Zhipu released GLM-5.3 on 14 August 2026, a post-training upgrade built on the same base model as GLM-5.2, with large gains on coding and cybersecurity benchmarks; NIST's CAISI later called it the most advanced open-weight model in cybersecurity.
Lovable raised a $400M Series C on 12 August 2026 at a $13.3B valuation, led by Menlo Ventures, after users created more than 60 million projects on the platform since its November 2024 launch.
On 12 August 2026 xAI released Grok 4.6, a month after Grok 4.5, aimed at long-running agents. Pricing stayed the same, and the model launched across Cursor, Grok Build and the API.
On 11 August 2026 Manus published a note saying it would soon operate as an independent company again, setting out when some users' data would be deleted and when access would return.
Alibaba released Qwen3.8-Max, a 2.4-trillion-parameter model, as a cloud service on 3 August 2026, and published open weights for a variant of it on 12 August 2026.
ByteDance released Seedance 2.5 in July 2026. The model generates up to 30 seconds of video per pass, offers a beta long-video mode, and can edit part of a scene without regenerating the whole clip. It is offered through an API.
On 24 July 2026 Anthropic released Claude Opus 5 at $5 and $25 per million tokens. Anthropic said it came close to Claude Fable 5's frontier intelligence at half the price.
On 23 July 2026 Black Forest Labs announced FLUX 3, a model family for images, video, audio and actions. The video model became generally available on 4 August 2026.
On 21 July 2026 Google announced three Gemini models at once: Gemini 3.6 Flash, a cheaper and faster 3.5 Flash-Lite, and 3.5 Flash Cyber, a security-focused variant limited to governments and trusted partners.
Moonshot AI released Kimi K3 on 16 July 2026, a model with 2.8 trillion total parameters. Its weights followed on 27 July 2026 under the Kimi K3 License.
On 16 July 2026 xAI released Grok 4.5, co-trained with Cursor, and made it the default model in Grok Build. xAI also published its coding benchmark results.
On 15 July 2026 Thinking Machines Lab released Inkling, an open-weight mixture-of-experts model with 975 billion total parameters. A smaller version followed on 31 July 2026.