Anthropic just rolled out a massive upgrade for its Claude Sonnet 4 AI model. Enterprise customers can now feed it prompts with up to 1 million tokens—around 750,000 words or 75,000 lines of code. That’s five times the old 200,000 token limit, and more than double the 400,000 tokens OpenAI’s GPT-5 offers.
This huge context window is live for API customers and on cloud platforms like Amazon Bedrock and Google Cloud’s Vertex AI.
Anthropic is pushing this to stay ahead as AI coding platforms grow and GPT-5 heats up competition. Claude powers tools like GitHub Copilot, Windsurf, and Cursor, but GPT-5’s pricing and coding chops are tempting some users away. Anysphere’s CEO even helped announce GPT-5, now Cursor’s default. 
Brad Abrams, Anthropic’s Claude product lead, downplayed any hit from GPT-5:
“I’m really happy with the API business and the way it’s been growing.”
Long context helps Claude nail complex coding tasks, especially ones that run for minutes or hours and require the AI to remember what it did earlier.
Anthropic also just updated Claude Opus 4.1 last week to boost coding capabilities further.
Other AI players are also pushing context limits: Google’s Gemini 2.5 Pro handles 2 million tokens, and Meta’s Llama 4 Scout claims 10 million. But studies hint that bigger isn’t always better—models often struggle to process all that info effectively. Abrams says Anthropic focused on boosting Claude’s “effective context window,” though he wouldn’t detail how.
Note: API pricing jumps for prompts over 200,000 tokens—input tokens now $6/million, output $22.50/million (up from $3 and $15).
This move signals Anthropic is doubling down on long-context AI to keep devs hooked—and its edge sharp.
