OpenAI rival Anthropic upgrades Claude, chatbot can now ingest entire codebases

Anthropic’s Claude 2.1 is possibly setting a new industry standard with a massive 200K token limit.

The open-source, universal standard is free to download by developers. (Image: Anthropic)

Listen to this article

00:00

1x 1.5x 1.8x

While OpenAI grapples with a CEO reshuffle, rival AI startup Anthropic has announced a sizable update for its own chatbot, Claude. On Tuesday, Anthropic unveiled Claude 2.1, which introduces new features like drastically increased token limits and reduced hallucination rates.

According to Anthropic, Claude Pro users can now process up to 200,000 tokens at once. For comparison, that’s over four times GPT-4’s limit of 32,000 tokens. Anthropic calls this an “industry first,” claiming it allows Claude to digest entire codebases in one go. The chatbot has also made “meaningful improvements” in comprehension and summarisation, particularly for long, complex documents like legal documents, financial reports, and technical specifications.

Beyond handling more data, Claude will hallucinate 50% less with this latest upgrade. Anthropic says this will result in more accurate, trustworthy responses. The chatbot can now pull data from APIs and external services to boost its knowledge, bringing it closer to ChatGPT’s functionality.

Story continues below this ad

Users can now also define what tools Claude should use through prompts, and the model has also gotten smarter at understanding what tool is best suited for a task – for instance, asking it to crunch some numbers may bring up the calculator.

Rounding out the update is new persistent memory and a test console for trying out prompts safely. With persistent memory, Claude can retain information like personalised instructions set by users. The test console lets developers tweak Claude’s responses before going live.

This comes just after ChatGPT announced an update that enables the ChatGPT voice feature for everyone. And in other news, OpenAI’s board approached Anthropic for a merger but got sent back.

Tags:
ChatGPT