AI model news, week of 12 October 2026
In the week of 12 October 2026, Together AI, xAI, OpenAI, Anthropic, DeepSeek, Google, and Mistral AI published 12 updates that matter for developers: new models (7), deprecations and retirements (1), and announcements (4).
New models
- Together adds serverless model reflection/Beam-1.0-501B-A23B
The new model has a context length of 1,048,576, NVFP4 quantization, function calling, structured outputs, and costs $0.40 per 1M input tokens and $1.20 per 1M output tokens.
- Meta Muse Spark Open is live on Together AI
Together AI now serves Muse Spark Open, Meta's open-weight model for agentic coding, with a 1M-token context window on serverless and dedicated infrastructure.
- New serverless model nvidia/NVIDIA-Nemotron-3.5-Super-VL with 262,144 context
Together now serves nvidia/NVIDIA-Nemotron-3.5-Super-VL serverless with 262,144 context, image input, function calling and structured outputs.
- Mistral introduces Timestral time-series foundation models
Mistral released Timestral, a family of four time-series foundation models with up to 209M parameters.
- xAI: Grok Voice 3.0 is available with Speech to Speech
grok-voice-3.0 is now available with Speech to Speech at the same price as Grok Voice 2.0, and the alias grok-voice-latest points to it.
- Grok Voice 1.0 is available with the Speech to Speech API
xAI's Grok Voice 1.0 is now available with the Speech to Speech API.
- Gemini 4 Argon (gemini-4-argon) is generally available in the Gemini API
Google released its most capable model, Gemini 4 Argon (gemini-4-argon), as generally available for complex coding, agentic, math, and science tasks.
Deprecations and retirements
- xAI deprecates grok-voice-2.0, upgrade to grok-voice-3.0 at the same price
xAI deprecated grok-voice-2.0 and directs developers to grok-voice-3.0 or grok-voice-latest at the same price.
Announcements
- DeepSeek opens up its in-house search API
DeepSeek opened up its in-house search API, powered by DeepSeek Search, which indexes hundreds of billions of web pages and supports multilingual search.
- OpenAI releases prompt-to-voice for all API customers
Developers can create a custom voice from a text description through the API or voice Playground and reuse it in GPT-Live sessions; nine new voices were also added.
- OpenAI adds native Decisions integration in GPT-Live
GPT-Live now supports native Decisions integration, letting apps act while the user is speaking and react to what Live says.
- Fast mode for Claude Opus 5.5 leaves research preview
Claude Opus 5.5 fast mode no longer needs an access request or the fast-mode-2026-02-01 beta header, and now has a default rate limit in every usage tier.
Models from this week on Zabio
| Model | Input, $/M | Output, $/M | Cached, $/M | Context window | 1,000 calls |
|---|---|---|---|---|---|
gemini-4-argonGoogle | 2.00 | 10.00 | no data | 200,000 | $9.00 |
Prices are the vendor list prices in USD per million tokens. "1,000 calls" is 1,000 requests of 2,000 input tokens and 500 output tokens each, without caching. Zabio bills the list price plus its service fee: see pricing.