Anthropic’s Claude Ai Model Can Now Handle Longer Prompts

Anthropic is expanding nan magnitude of accusation that endeavor customers tin nonstop to Claude successful a azygous prompt, portion of an effort to pull much developers to nan company’s celebrated AI coding models.

For Anthropic’s API customers, nan company’s Claude Sonnet 4 AI exemplary now has a 1 cardinal token discourse model — meaning nan AI tin grip requests arsenic agelong arsenic 750,000 words, much than nan full Lord of nan Rings trilogy, aliases 75,000 lines of code. That’s astir 5 times Claude’s erstwhile limit (200,000 tokens), and much than double nan 400,000 token discourse model offered by OpenAI’s GPT-5.

Long discourse will besides beryllium disposable for Claude Sonnet 4 done Anthropic’s unreality partners, including connected Amazon Bedrock and Google Cloud’s Vertex AI.

Anthropic has built 1 of nan largest endeavor businesses among AI exemplary developers, mostly by trading Claude to AI coding platforms specified arsenic Microsoft’s GitHub Copilot, Windsurf, and Anysphere’s Cursor. While Claude has go nan exemplary of prime among developers, GPT-5 whitethorn frighten Anthropic’s power pinch its competitory pricing and beardown coding performance. Anysphere CEO Michael Truell moreover helped OpenAI denote nan launch of GPT-5, which is now nan default AI exemplary for caller users successful Cursor.

Anthropic’s merchandise lead for nan Claude platform, Brad Abrams, told TechCrunch successful an question and reply that he expects AI coding platforms to get a “lot of benefit” from this update. When asked if GPT-5 put a dent successful Claude’s API usage, Abrams downplayed nan concern, saying he’s “really happy pinch nan API business and nan measurement it’s been growing.”

Whereas OpenAI generates astir of its gross from user subscriptions to ChatGPT, Anthropic’s business centers astir trading AI models to enterprises done an API. That’s made AI coding platforms a cardinal customer for Anthropic, and could beryllium why nan institution is throwing successful immoderate caller perks to pull users successful nan look of GPT-5.

Last week, Anthropic unveiled an updated type of its largest AI model, Claude Opus 4.1, which pushed nan company’s AI coding capabilities a spot further.

Generally speaking, AI models thin to execute amended connected each tasks erstwhile they person much context, but particularly for package engineering problems. For example, if you inquire an AI exemplary to rotation up a caller characteristic for your app, it’s apt to do a amended occupation if it tin spot nan full project, alternatively than conscionable a mini section.

Abrams besides told TechCrunch that Claude’s ample discourse model helps it execute amended astatine agelong agentic coding tasks, successful which nan AI exemplary is autonomously moving connected a problem for minutes aliases hours. With a ample discourse window, Claude tin retrieve each its erstwhile steps successful long-horizon tasks.

But immoderate companies person taken ample discourse windows to an extreme, claiming their AI models tin process monolithic prompts. Google offers a 2 cardinal token discourse model for Gemini 2.5 Pro, and Meta offers a 10 cardinal token discourse model for Llama 4 Scout.

Some studies propose there’s a limit to really ample discourse windows tin be, and AI models are not great astatine processing monolithic prompts. Abrams said that Anthropic’s investigation squad focused connected expanding not conscionable nan discourse model for Claude, but nan “effective discourse window,” suggesting that its AI tin understand astir of nan accusation it’s given. However, he declined to uncover Anthropic’s nonstop techniques.

When prompts to Claude Sonnet 4 are complete 200,000 tokens, Anthropic will complaint much to API users, astatine $6 per cardinal input tokens and $22.50 per cardinal output tokens (up from $3 per cardinal input tokens and $15 per cardinal output tokens).