AIStall
EditionBy The AI Stall editorsOctober 10, 2026

Three Things in AI: Haiku 5.5 Pricing and Two Free Upgrades

Three Things in AI: Haiku 5.5 Pricing and Two Free Upgrades

Three things, read from the vendors' own pages on October 10, 2026. One is a price cut on a small model that most readers will meet inside other products rather than directly. One is a free safety fence for anyone who lets a coding agent run commands on a real laptop. One is a quiet upgrade to Google's automation builder for every Workspace customer who already has it. Same three slots as issue zero promised: what shipped, what it costs, who it changes the week for. Coffee first. ☕

What this note covers, in order.
What this note covers, in order.

One: Anthropic prices Haiku 5.5 at ten cents per million input tokens

What shipped. On October 7, 2026 Anthropic released Claude Haiku 5.5, described on [the announcement page](https://www.anthropic.com/claude-haiku-5-5) as its cheapest, fastest and most capable small model to date. The page positions it for summaries, classification, database queries and live customer support, and as a subagent next to Opus 5.5 and Sonnet 5.5 for coding work.

It is available on the Claude Platform API under the model name claude-haiku-5-5, and on Amazon Bedrock, Google Cloud Vertex AI and Microsoft Azure, per the same page. The page does not state a context window, so this brief does not either.

What it costs. Per [the Claude pricing page](https://claude.com/pricing), checked October 10, 2026: $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Above 100,000 tokens the rate is $0.50 and $2.50. Haiku 4.5 is still listed at $1 and $5, so the short-prompt rate is 90 percent lower, which matches the footnote on the announcement.

Who it changes the week for. Directly, teams with a developer who already calls a model API for high-volume jobs: ticket tagging, summaries, data extraction. Indirectly, everyone else, because the same model landed in GitHub Copilot the same day for Pro, Pro+, Max, Business and Enterprise plans, per [the Copilot changelog](https://github.blog/changelog/2026-10-07-claude-haiku-5-5-in-github-copilot).

One caution from the vendor itself: the announcement says the new tokenizer uses slightly more tokens per task than the old one. The saving per job is therefore a little smaller than the saving per token. Anthropic's own average is about 75 percent less to run, not 90.

Two: GitHub puts a local sandbox around Copilot's agent, at no extra cost

What shipped. On October 7, 2026 GitHub's changelog announced that [local sandboxing for GitHub Copilot is generally available](https://github.blog/changelog/2026-10-07-local-sandboxing-for-github-copilot-now-generally-available). When the agent runs a command or a tool on a developer's machine, the sandbox limits what that command can read, write, reach on the network, and borrow in the way of Git and GitHub credentials.

The policies are set by the developer or the organisation. Enterprise-managed settings can make sandboxing mandatory, so an individual cannot loosen it. It works in Copilot CLI, the Copilot app and VS Code sessions using Agent Host, on Windows, macOS and Linux, and the page says the policy applies whichever model the agent is using.

What it costs. The changelog says local sandboxing is included with GitHub Copilot at no additional cost. It does not name plan tiers and it does not give setup steps; both live in the documentation it links to.

Who it changes the week for. Any small team that has let a coding agent loose on a real laptop and later wondered what else it touched.

The practical change is that "the agent may run commands" and "the agent may read my keys and call the internet" are no longer the same permission. Whoever owns the GitHub organisation should switch the requirement on this week; the developers will not notice until it stops something.

Three: Workspace Studio learns nested conditions and a Drive search step

What shipped. On October 9, 2026 Google's Workspace Updates blog described [new logic and search steps in Workspace Studio](http://workspaceupdates.googleblog.com/2026/10/new-logic-and-search-steps-in-workspace-Studio-help-expand-automation-capabilities.html), Google's built-in automation builder. The additions are nested conditionals for multi-layer branching, a skip-replies option for Gmail thread triggers, fuller Meet and Calendar outputs including complete invitee lists and follow-up email drafts, and a Drive step that searches for a document.

Rollout starts October 13, 2026 and can take up to 15 days to become visible, per the post. The features are on by default for end users, with no admin switch mentioned.

What it costs. Nothing extra. The post lists availability as all Google Workspace customers and Workspace Individual subscribers who have Workspace Studio access. If your edition does not already include Studio, this changes nothing for you.

Who it changes the week for. Teams that run their small automations inside Google rather than in a separate tool.

Nested conditions are the difference between "if the email is from a customer, do X" and "if it is from a customer, and the subject mentions an invoice, and nobody has replied yet, do X".

The skip-replies option alone removes the most common way these automations misfire, which is firing again on every reply in the thread.

What it costs, in one place

Not in today's three, and why

Three things were close. 5, released September 28, 2026 at $2 and $10 per million tokens, is real and useful, but it is twelve days old and a morning brief should not pretend otherwise; the pricing page above already carries it.

com and PandaDoc inside the Gemini chat, is held for a later edition so the list can be printed in full.

Microsoft's October 8 post about CRM inside Teams and a Sales Development Agent in public preview is tomorrow's lead item, because the part that matters, the agent starting to consume Copilot credits on November 16, 2026, deserves a slot rather than a line.

OpenAI's news page returned an error when read on October 10, 2026; the rule from issue zero is primary source or nothing.

Bottom line

5 exists at a tenth of the old Haiku price for short prompts, and that the Copilot sandbox is live, free, and should be required at organisation level rather than left to each laptop.

If you run on Google Workspace with Studio: open your automations after October 13 and add a skip-replies condition to anything triggered by Gmail.

If neither applies, today cost you three minutes and nothing in your week changed, which is a legitimate outcome. Wednesday's edition covered stacked pull requests and Slack To-dos if you missed it. Tomorrow, three more.

Sources