News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
Claude 2

Claude 2

Anthropic's 2023 assistant model with a 100K-token window and stronger coding.

Claude 2 is a general-purpose language model from Anthropic, released in 2023 as the successor to Claude 1.3. It handles long inputs, writes extended documents, and improved on coding, maths and reasoning tests, powering both the claude.ai chat beta and a business API.

Share on WhatsApp

  • Accepts up to 100K tokens in a single prompt
  • 76.5% on the multiple-choice Bar exam, up from 73.0%
  • 71.2% on Codex HumanEval, up from 56.0% for Claude 1.3
  • 88.0% on GSM8k grade-school maths problems
  • Harmless responses twice as often as Claude 1.3 in internal red-teaming

What it is

Claude 2 is a language model that Anthropic built to work as a conversational helper and general assistant. It arrived in 2023, following the company's earlier Claude 1.3. The company positions it as a colleague you instruct in plain language for everyday work. Compared with its predecessor it handles longer exchanges, produces lengthier written output, and does better on coding, maths and reasoning tasks. It reached the public through two routes: a business API priced the same as Claude 1.3, and a beta chat site at claude.ai open to people in the US and UK.

How it came about

Anthropic announced Claude 2 as its newest model, describing gains over earlier Claude versions in coding, maths and reasoning. The company said it had raised both input and output lengths while working on performance and safety together. The chat experience launched as an open beta, and the firm said it was already working with thousands of businesses on the API. Partners named at launch include Jasper, a content platform, and Sourcegraph, whose coding assistant Cody draws on the model. Wider country coverage was described as work in progress.

Inside the model

Anthropic gives no layer counts, parameter totals or training compute for Claude 2. What it does say is that each prompt may carry up to 100K tokens, enough to read hundreds of pages of technical documentation or a full book, and that replies can run to a few thousand tokens in one pass. Training data is more recent than the previous version, so the model recognises newer frameworks and libraries. On the safety side the company reports an internal red-teaming evaluation built from a large representative set of harmful prompts, scored automatically and checked by hand, plus a range of other safety techniques.

What it can do

Anthropic reports several test results. Claude 2 reached 76.5% on the multiple-choice Bar exam, against 73.0% for Claude 1.3. On Codex HumanEval, a Python coding test, it scored 71.2% versus 56.0% before. GSM8k, a set of grade-school maths problems, came in at 88.0% against 85.2%. For the GRE it sits above the 90th percentile in reading and writing and near the median for quantitative reasoning among graduate school applicants. The company also notes that no model is immune to jailbreaks and that inappropriate output remains possible.

How to use it

Two routes were offered. Businesses could use the Claude 2 API at the same price as Claude 1.3. Individuals in the US and UK could sign up for the claude.ai chat beta and talk to the model in natural language. Anthropic said it was working toward broader geographic availability and invited people in unsupported regions to get in touch.

Safety and testing

Anthropic says it iterated on Claude 2 to make it harder to steer toward offensive or dangerous output. Its internal red-teaming evaluation covers a large representative set of harmful prompts, run as an automated test with regular manual review, and the model gave harmless responses twice as often as Claude 1.3 on that measure. The company adds that jailbreaks remain possible despite the techniques and red-teaming applied, and that the chat launch is an open beta where inappropriate responses can still occur.

Reception and debate

Partner reactions were positive. Jasper's engineering leadership praised the model's semantics, reasoning on complex prompts and a context window three times larger than before, and said it competed well across many use cases, with an edge in long-form, low-latency work. Sourcegraph's founders highlighted the reasoning quality and the 100K window for passing codebase context to Cody. Anthropic itself framed the chat launch as a beta and asked users for feedback as it rolled the product out more widely.

Why it matters

Claude 2 extended the Claude line with much longer inputs and stronger coding and maths, and it was the model behind Anthropic's first public chat beta at claude.ai. The company reported thousands of businesses already working with the API, and named Jasper and Sourcegraph as early partners building products on it. Improvements were promised to ship gradually over the following months.

Questions people ask about Claude 2

Who made Claude 2 and when did it appear?

Anthropic released it in 2023 as its newest model, succeeding Claude 1.3. It was offered through a business API and, for people in the US and UK, a beta chat site at claude.ai.

How much text can it handle at once?

Each prompt may contain up to 100K tokens, which Anthropic says is enough for hundreds of pages of technical documentation or an entire book. Replies can also run to a few thousand tokens in a single go.

What changed compared with Claude 1.3?

Anthropic reports gains in coding, maths and reasoning, longer inputs and outputs, and more recent training data. On its internal harmful-prompt test the model gave harmless answers twice as often as the earlier version.

How do you get access to it?

Businesses could call the API at the same price as Claude 1.3. Individuals in the US and UK could register for the claude.ai chat beta. Anthropic said wider availability was being prepared.

Are there limits on how it should be used?

Anthropic describes the chat launch as an open beta and warns that inappropriate responses can still happen. It points to everyday uses such as summarising or organising information, and notes that jailbreaks remain possible.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Claude 2 yet, so the facts on this page come from the sources listed above: anthropic.com/news/claude-2. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.