Today, August 26, OpenAI retires two products at once. The o3 reasoning model disappears from ChatGPT’s model picker — the last standalone reasoning flagship from the pre-GPT-5 era. And the Assistants API, a server-side infrastructure layer running enterprise chatbots and agent workflows since November 2023, returns hard errors instead of responses for any application still calling it.
The o3 exit is the one getting attention. The Assistants API shutdown is the one that will break production systems.
o3: the end of standalone reasoning
OpenAI announced the 90-day sunset on May 28, giving o3 a clean calendar endpoint. The model launched in April 2025 as the flagship of the o-series — trained via reinforcement learning to generate internal reasoning steps before producing a visible answer. It benchmarked at 96.7 per cent on AIME 2024, 87.7 per cent on GPQA Diamond, and a 2727 Codeforces Elo. For a specific population of power users — competitive mathematics, multi-hop code debugging, scientific literature analysis — o3’s reasoning behaviour differed from standard language models in ways that mattered.
It has not been the default option in ChatGPT for months. OpenAI says only a fraction of a per cent of daily users still manually select it. But for those who do, today closes a chapter.
One carve-out: o3-pro, the higher-compute variant, remains available in ChatGPT for Pro, Team, Enterprise, and Edu subscribers after today. The o3 API stays accessible through separate developer endpoints until December 11, 2026.
The replacement is GPT-5.6 Sol. For teams doing high-volume reasoning work, OpenAI’s migration guides point to GPT-5.6 Terra — a mid-tier option in the same family — at roughly 40 per cent of Sol’s cost. The andrew.ooo migration analysis recommends Terra as the default for most o3 workloads at scale. This continues a retirement pattern we tracked when GPT-5.6 launched in July — the same model family that replaced o3 is now itself the only option.
The Assistants API: the deadline that actually breaks things
The Assistants API was a three-layer server-side system: Assistants (persistent objects bundling a model, instructions, and tools), Threads (server-side conversation stores held by OpenAI indefinitely), and Runs (asynchronous execution processes). Since November 2023, it has been handling conversation state, tool execution, and task orchestration for enterprise applications — customer service bots, multi-session AI agents, automated workflows.
After today, calls to /v1/assistants, /v1/threads, or /v1/runs do not return a deprecation warning or a rate limit. They return a hard error. According to TechTimes’ analysis, OpenAI has provided no grace period, no extension option, and no automated migration. Threads — the full conversation histories stored server-side — do not transfer automatically. Developers who have not built their own state management and migrated their agents to the Responses API will lose that state.
The scale of OpenAI’s 2026 retirement cycle is worth noting. The company has retired GPT-4o, GPT-4.1, o4-mini, GPT-4.5, and now o3 from ChatGPT — more models in one year than in all prior years combined. The model lifecycle has compressed from roughly 18 months to closer to six.
The pattern behind the closures
What stands out is the speed. o3 launched in April 2025. Sixteen months later, it is gone from the consumer product. The Assistants API launched in November 2023. Thirty-three months later, it is gone entirely. OpenAI is compressing its own product lifecycles faster than enterprise software norms — and faster than many of its enterprise customers’ upgrade cycles.
This is not unusual for a platform racing to ship. It is unusual for the enterprises that built on the assumption of stability. The lesson, increasingly clear across the AI industry: if you build on an API, own your state and abstract your model calls. The model you chose last year may not exist next quarter.
NZ angle
For New Zealand developers using OpenAI’s stack — and there are more than a few, from Wellington-based AI agencies to Auckland fintechs — the Assistants API shutdown is the more pressing deadline. Any local team that built a customer service bot or multi-session agent on the Assistants API in 2024 or early 2025 needs to have migrated to the Responses API by today or face broken production systems. The official migration guide is the starting point.
❓ FAQ
What is OpenAI o3? A reasoning model released in April 2025 that generated internal reasoning steps before answering. It was designed for complex mathematics, code debugging, and scientific analysis. GPT-5.6 Sol and Terra are its successors.
What was the Assistants API? A server-side infrastructure layer that handled conversation state, tool execution, and task orchestration for developers building AI agents and chatbots. It launched in November 2023 and is being replaced by the Responses API.
Does the o3 API still work? Yes. The o3 API remains accessible through developer endpoints until December 11, 2026. Only the ChatGPT consumer interface loses o3 today.
What should developers do? For o3 workloads, migrate to GPT-5.6 Terra (cost-effective) or Sol (flagship). For Assistants API applications, migrate to the Responses API and implement local conversation state management. There is no grace period.
