Claude Instant

What it was

Claude Instant was a version of Anthropic's Claude language model built to be lighter, less expensive and much faster than the full Claude model. Anthropic announced it on 14 March 2023, the same day it introduced Claude itself.^[1]

The reasoning behind a model like this is straightforward: the largest model is not always the right tool. For high-volume, comparatively simple work — sorting incoming messages, pulling a field out of a document, answering a routine question — what matters is how quickly and how cheaply a system responds, not whether it can handle the hardest reasoning task in the room. A smaller model that answers in a fraction of the time and at a fraction of the cost is often the better engineering choice, and it lets a company reserve the expensive model for the queries that actually need it.

Release and access

Claude Instant was released for business use through Anthropic's API rather than as a stand-alone consumer product. Among the partners at launch was Quora, whose Poe app offered both Claude and Claude Instant.^[1] The API eventually carried three versions of the model, labelled claude-instant-1.0, claude-instant-1.1 and claude-instant-1.2.^[3]

Claude Instant 1.2

The third version, released on 9 August 2023, brought strengths of the larger Claude 2 model into the smaller, faster one.^[2] Anthropic reported two benchmark results alongside it. On Anthropic's Codex coding evaluation the model scored 58.7%, against 52.8% for Instant 1.1; on GSM8K, a set of grade-school mathematics word problems used to test multi-step arithmetic, it scored 86.7%, against 80.9%.^[2] Both measure how often the model lands on an answer that can be checked for correctness.

Anthropic also said that Instant 1.2 hallucinated less and was more resistant to jailbreaks than Instant 1.1, a judgement based on automated red-teaming — testing in which a system deliberately tries to talk a model past its own safeguards. A hallucination is a confident statement the model has no real basis for; a jailbreak is a prompt designed to get the model to drop its instructions.^[2] Instant 1.2 was offered through Anthropic's API, with businesses able to contact sales.^[2]

Place in the lineup

Anthropic's next fast, low-cost model was Claude 3 Haiku, released in March 2024 as part of the Claude 3 family.^[4] The name changed, but the role carried over: beneath the flagship model there is always a need for something small, quick and inexpensive.

The three numbered versions also show how a model family is maintained in practice. Between the first release and the last, the changes were measured in benchmark points and in safety claims rather than in a new architecture, and each version replaced the one before it inside the same API. Customers who had built around claude-instant-1.0 could move to 1.1 and then to 1.2 without changing what their product did.

Retirement

On 4 September 2024 Anthropic notified developers that all Claude Instant models would be retired, and the models were shut off on 6 November 2024.^[3] Anthropic's deprecation table listed claude-haiku-4-5-20251001 as the replacement for claude-instant-1.0, 1.1 and 1.2 alike.^[3] With that, the cheap-and-fast slot in Anthropic's offering passed entirely to the Haiku line, and Claude Instant became a historical entry in the API documentation rather than a model anyone could call.