Claude Sonnet 4

Retired Jun 2026

Released May 2025 alongside Opus 4, with coding and reasoning upgrades over Sonnet 3.7 and hybrid instant and extended thinking.

Context: 200K tokens
Max output: 64K
Cutoff: Mar 2025
Released: May 2025

Visit Anthropic Compare vs other LLMs →

Lyceum

NVIDIA H200 from $2.20/hr, B200 from $3.40/hr, B300 from $3.80/hr

Per-second billing with no minimum commitment or vendor lock-in

Scale to 8,000 GPUs over InfiniBand for distributed AI training

View Pricing

Capabilities

What Claude Sonnet 4 accepts and can do, as published by Anthropic.

Text Accepts and generates natural-language text.
Image Accepts images as input alongside text.
Reasoning Thinks before it answers, always on or as a switchable mode.
Function calling Connect to external tools, APIs, and systems.
Structured output Return responses in structured formats like JSON.

Common questions

What is Claude Sonnet 4 good for?

Coding and reasoning, with a choice of instant or extended thinking. It takes text and images, at the Sonnet tier below Opus.

When is Claude Sonnet 4 not a good fit?

Any new work: Anthropic retired it in June 2026, and Sonnet 4.5 and 4.6 know more recent events. No audio or video input, and closed weights, so no self-hosting.

How does Claude Sonnet 4 compare with Claude Sonnet 4.5?

Anthropic says Claude Sonnet 4.5 improves on Claude Sonnet 4 most in computer use, reasoning and math, and holds focus on long multi-step tasks for many hours. The launch added a memory tool and context editing so agents can run longer. Same 200K-token context and text and image input. Claude Sonnet 4 came out 4 months earlier and has a newer knowledge cutoff (March 2025 against January 2025).

Can I self-host Claude Sonnet 4?

No. Anthropic does not publish the weights; Claude Sonnet 4 is available only through hosted APIs.