In October 2026 the AI labs focused on two things: cheaper, faster models for high-volume work, and agents that carry out tasks themselves in browsers and apps. These are the main updates confirmed by their official sources.
Google: Gemini 4 Argon is generally available
- October 13: Gemini 4 Argon reached general availability in the Gemini API, aimed at coding, agentic tasks, math and science.
- October 30: the Computer Use tool became generally available, so a model can work through real interfaces by clicking and typing.
- October 6: the Nano Banana 2.1 image model became generally available for generating and editing images.
- Older models such as gemini-3.5-flash and gemini-3.7-flash were deprecated and routed automatically to newer ones.
Anthropic: Claude Haiku 5.5 (October 7)
Anthropic launched Claude Haiku 5.5, which it calls its cheapest, fastest and most capable small model yet. It is built for high-volume, cost-sensitive work such as summaries, classification and live customer support, and as a subagent next to larger models.
- About 75% cheaper on average than Haiku 4.5.
- $0.10 per million input tokens and $0.50 per million output tokens (prompts up to 100k tokens).
- The first Haiku-class model with an adjustable effort setting to trade cost against intelligence.
- Available on the Claude Platform and on AWS, Google Cloud and Microsoft Azure.
On October 22, Anthropic also donated its open Agent Skills standard to the Agentic AI Foundation, making it a shared standard for teaching agents specific skills.
What it means for your project
- A customer-service bot on WhatsApp or your site is now much cheaper to run on the new small models, without lower answer quality.
- Automating repetitive work inside systems (data entry, request review) is practical with officially supported computer-use tools.
- If your project relies on an older Gemini model, check the deprecation list before your bot's behaviour changes unexpectedly.