Week of 5 October 2026: cheaper small models and agents move into everyday tools

The models and assistants businesses build on had a busy week. Anthropic cut small-model prices, ChatGPT changed its default model and Google showed an agent for work. OpenAI also began watermarking text for the EU, and app teams got a new iPhone screen to plan for.

Anthropic cuts small-model prices with Claude Haiku 5.5

On 7 October Anthropic released Claude Haiku 5.5 (model ID claude-haiku-5-5) on its own platform, AWS, Google Cloud and Microsoft Azure. Prompts up to 100,000 tokens cost $0.10 per million input tokens and $0.50 per million output tokens, against $1 and $5 for Haiku 4.5, while longer prompts cost $0.50 and $2.50. Anthropic puts the typical saving near 75%, a figure that already allows for a new tokenizer that counts slightly more tokens for the same work. The same release halved Sonnet 5.5 cache reads from $0.20 to $0.10 per million tokens. VentureBeat pointed out that the lower Haiku rates equal those of OpenAI’s GPT-6 Luna, although Luna’s higher rate only starts above 272,000 input tokens.

Why it matters: Classification, summarizing and routing steps in existing pipelines are good candidates to move to the cheaper model. Re-check the cost on real prompts first, because the tokenizer change and the 100,000-token threshold both change the bill.

Sources: Introducing Claude Haiku 5.5 (Anthropic, 7 Oct 2026), Anthropic launches Claude Haiku 5.5 with 90% API price reduction, matching GPT-6 Luna (VentureBeat, 7 Oct 2026).

OpenAI moves ChatGPT to GPT-6 and cuts its API tiers to three

On 7 October OpenAI began moving ChatGPT Plus, Pro, Business and Enterprise users to GPT-6 Sol, and Free and Go users started getting GPT-6 Luna on 8 October. The release also brought Intelligent UI, which lets answers include charts, forms and small interactive tools such as calculators. OpenAI’s system card says Codex and ChatGPT Work stay on the September model versions for now. A day earlier, on 6 October, OpenAI cut its paid API usage tiers from five to three (Build, Launch and Grow). Grow, the new top tier, is reached after $500 in cumulative API payments instead of $1,000, and organizations already on a paid tier were reassigned without having to do anything.

Why it matters: Teams that use ChatGPT for repeated internal tasks should spot-check results after the model change. Smaller API accounts can now reach the top usage tier with less total spend.

Sources: GPT-6 and Intelligent UI in ChatGPT (OpenAI Developer Community, 7 Oct 2026), GPT-6 Sol and GPT-6 Luna: October 2026 update (OpenAI Deployment Safety Hub, 7 Oct 2026), Update to OpenAI API Rate Limits (OpenAI Developer Community, 6 Oct 2026), OpenAI brings GPT-6 to ChatGPT and debuts Intelligent UI (9to5Mac, 7 Oct 2026).

Google previews a single Gemini agent for work

At Gemini at Work 2026 on 8 October, Google Cloud announced the Gemini agent. Given a goal, it plans the steps itself and works across Gmail, Docs, Sheets and Chat as well as Slack and Microsoft 365, and an API lets other apps run it headless. It picks a model for each task, currently from the Gemini and Claude families. Reworked reports that each agent gets its own identity and action log, admins can set a hard spending cap per project and ready-made connectors include Salesforce, BigQuery, Snowflake and Postgres. 9to5Google says the agent is in private preview, with wider availability planned soon for select Workspace Business and Enterprise plans.

Why it matters: Companies on Google Workspace should decide which systems the agent may reach and set spending limits before it arrives on their plan, not after staff start using it.

Sources: Google Cloud introduces the Gemini agent (Google, 8 Oct 2026), Google Cloud announces ‘Gemini agent’ as ‘universal agent for work’ (9to5Google, 8 Oct 2026), Google Debuts Gemini, an All-Purpose Agent for Work (Reworked, 9 Oct 2026).

OpenAI starts watermarking AI text for the EU AI Act

On 5 October OpenAI introduced textGrain, a statistical watermark carried in the word choices its models make, as part of its plan for meeting the EU AI Act’s text provenance rules. API customers anywhere can turn it on for selected models (it is off by default), and eligible ChatGPT and Codex text produced in the EU will carry it over the following weeks. At first, only approved researchers and organizations can use the detector. The Register reported that detection falls off quickly once text is edited: replacing a tenth of the words in a 400-token passage with synonyms cut the detection rate from 92% to 66%.

Why it matters: Businesses that publish or send AI-drafted text in the EU should not treat a watermark, or the lack of one, as proof of who wrote it, and should keep their own record of where AI was used.

Sources: OpenAI’s approach to EU text provenance rules (OpenAI Developer Community, 5 Oct 2026), OpenAI rolls out weak sauce watermarking for AI text (The Register, 6 Oct 2026).

Apple opens App Store submissions for iPhone Duo

On 5 October Apple began accepting apps optimized for iPhone Duo in App Store Connect, ahead of the device reaching customers on 23 October. Rebuilding with Xcode 27.1 adds support, and Device Hub previews an app in each pose and orientation the device can take. Existing apps still run without changes, but MacRumors reports that only apps built with the iOS 27.1 SDK or later fill the inner display, and the rest get black borders. Uploading iPhone Duo screenshots adds an optimized badge to an app’s product page, and from April 2027 Apple will require those screenshots for every app and game submitted.

Why it matters: Any company with an iOS app should plan a rebuild and new screenshots now, because from April 2027 every app submission will need iPhone Duo screenshots.

Sources: Prepare and submit your apps for iPhone Duo (Apple Developer, 5 Oct 2026), Apple Announces iPhone Duo Apps Can Now Be Submitted to App Store (MacRumors, 5 Oct 2026), Apple invites developers to submit iPhone Duo-ready apps to the App Store (9to5Mac, 5 Oct 2026).

This digest summarizes reporting from the linked sources in our own words. Follow the links for the full stories.