AI news
What changed in AI this month, in plain language. Every item links to the maker’s own announcement. See the full AI model tracker.
OpenAI adds prompt-to-voice and nine new voices to GPT-Live in the API
OpenAI's API changelog says developers can now create a custom voice from a text description, plus nine new GPT-Live voices and Decisions integration.
Text modelsOpenAI adds Ultrafast mode for GPT-6.1 Sol in the Responses API
OpenAI says Ultrafast mode for GPT-6.1 Sol shortens the time between generated tokens and is open to all API users, subject to rate limits.
Text modelsAnthropic launches Claude Haiku 5.5, its cheapest and fastest small model
Claude Haiku 5.5 launched October 7, 2026 with an adjustable effort setting and lower prices; Anthropic also halved Sonnet 5.5 cache-read prices.
Image modelsMidjourney tests a Thinking Mode that reruns images to fix prompt misses
Midjourney is testing Thinking Mode on its alpha site: rerun a V8.2 image and it looks for where the result missed your prompt, then generates again.
Text models · Open weightsMistral Large 4 opens in public preview, with open weights promised by month end
Mistral AI released a public preview of Mistral Large 4, a 1 trillion parameter multimodal model, and says the open weights will follow by the end of October.
Image modelsGoogle publishes the Nano Banana 2.1 model card
Google DeepMind published the model card for Nano Banana 2.1 on October 6, 2026: an image model built on Gemini 3.6 Flash with 4K output.
Image modelsBlack Forest Labs launches FLUX 3 Image with bounding boxes and ten references
FLUX 3 Image generates and edits from one endpoint, lets you place elements with bounding boxes, combines up to ten references and outputs up to 4K.
Text modelsGoogle announces Gemini 4 Argon, rolling out first to cyber defenders
Google DeepMind introduced Gemini 4 Argon, a new frontier model, and is releasing it first to trusted cyber defenders through its Fairwind Program.
Text modelsOpenAI releases GPT-6.1 Sol at a fraction of GPT-6 Astra's price
OpenAI says GPT-6.1 Sol nearly matches GPT-6 Astra on agentic coding and professional work at one-fifth of the standard token prices.
Text modelsAnthropic launches Claude Sonnet 5.5, faster and cheaper than Sonnet 5
Claude Sonnet 5.5 is the second model in the Claude 5.5 family, priced at $2 and $10 per million tokens, with Haiku 5.5 due in the following weeks.
Music and voiceElevenLabs releases Eleven v4 and a low-latency v4 Turbo
ElevenLabs launched Eleven v4, its most expressive text-to-speech model, and Eleven v4 Turbo for fast voice agents, with support for 90+ languages.
Video modelsKling 4.0 Flash opens to early-access users ahead of a wider October rollout
Kling began letting a limited group try Kling 4.0 Flash on September 28, with a wider rollout of Kling 4.0 planned for October 2026.
Video modelsOpenAI's Sora video models leave the API on September 24
OpenAI removed the Sora 2 models and the Videos API on September 24, 2026, after closing the Sora app and website on April 26.
Text models · Open weightsDeepSeek releases V4.1-Flash and begins phasing out V4-Pro
DeepSeek's V4.1-Flash is a 552B-parameter model with native multimodal support and lower prices, and V4-Pro requests will route to it.
Music and voiceSuno releases v6, a music model generation built with Warner Music Group and BMG
Suno v6 arrives as v6, v6-wild and the faster v6-mini, developed with industry partners including Warner Music Group, BMG and Believe.
Image modelsMidjourney adds an Edit Model for V8 that replaces Omni Reference
Midjourney's Edit Model edits images with instructions, uses up to four image references, and supports inpainting and outpainting in V8.2.