Five AI deadlines in August 2026: retirements and price hikes

Nobody sends you a calendar invite when your AI vendor retires a model. They post it to a docs page you don't read and email an address that forwards to a shared inbox. So here are five AI deadlines in August 2026 we pulled straight from vendor documentation — four retirements and one price increase, all dated, all capable of breaking something you shipped months ago.

What actually happened

August 4 — Gemini 3.5 Flash leaves the global region. Per Google Cloud's Gemini Enterprise release notes, Gemini 3.5 Flash "will be removed as a model from the global region in the Gemini Enterprise app on August 4, 2026." That's tomorrow.

August 5 — Claude Opus 4.1 retires. Anthropic's deprecation page lists claude-opus-4-1-20250805 with a retirement date of August 5, 2026, replacement claude-opus-4-8. Anthropic's own wording on retired models: requests will fail. Note the dates apply to the Claude API, Claude Platform on AWS, and Microsoft Foundry — Bedrock and Google Cloud run their own schedules, so "it still works on Bedrock" is not evidence you're fine.

August 26 — OpenAI's Assistants API shuts down. OpenAI's deprecations page puts the Assistants API shutdown at 2026-08-26, with the Responses API and Conversations API as the migration path. This is the heaviest lift on the list: it's an architecture change, not a string swap. (ChatGPT Atlas also goes dark this month — we covered that one when it was announced.)

August 31 — Moonshot sunsets kimi-k2.5 and moonshot-v1. The Kimi platform model list states both are already closed to newly registered users, with "full platform sunset on August 31." Recommended replacement is kimi-k3.

August 31 — Claude Sonnet 5's introductory pricing ends. Anthropic's pricing page is explicit: $2/$10 per million input/output tokens through August 31, then $3/$15 from September 1. Batch pricing moves the same way, $1/$5 to $1.50/$7.50. That's a flat 50% increase on a model a lot of people quietly made their default.

Why AI model deadlines matter for your business

Four of these five are one-line changes if your model choice lives in one place. If it doesn't — if a model string is scattered across a Lambda, a cron job, a Zapier step, and a prototype someone shipped to production in March — then a vendor's routine housekeeping becomes your Tuesday incident. We wrote the same warning when DeepSeek killed its aliases in July. The names change; the failure mode doesn't.

The pricing one is different and worth its own line in your forecast. A 50% step-up on Sonnet 5 doesn't announce itself — it shows up as a September invoice that's half again what August was, on identical usage. If you sized an automation's ROI at $2/$10, resize it at $3/$15 now and decide deliberately whether it still clears the bar, or whether that workload belongs on Haiku.

Practical version: grep your repos for every model string today, write down which ones appear more than once, and put the September number in your budget before September prints it for you.

Sources: Google Cloud Gemini Enterprise release notes, Anthropic model deprecations, Anthropic pricing, OpenAI deprecations, Kimi platform model list.