← Back to API models

Claude Sonnet 5

by Anthropic · Sonnet flagship · released Jun 30, 2026

Near-Opus quality at a permanent $2/$10 — Anthropic cancelled the planned September price increase. The default Claude for most production work.

Input $2 / 1M
Output $10 / 1M
Context 1M
Pricing Permanent (was intro)
Anthropic Console ↗ Updated August 17, 2026
§ API pricing

Per-token rates — now permanent.

Input
$2/1M tokens
Prompt tokens
  • Made permanent Aug 10, 2026
  • Vision inputs billed as tokens
  • Prompt caching available
Output
$10/1M tokens
Completion tokens
  • Planned Sep 1 rise to $15 cancelled
  • Cheaper than GPT-5.4's $15 output
  • Includes extended thinking tokens
Context
1Mtokens
Window
  • 1M context window
  • Beta tier may have separate pricing
  • Largest Claude context available
Caching
~90%discount on cache hits
Prompt caching
  • Reuse large system prompts cheaply
  • Big wins for chat with long instructions
  • 5-min and 1-hour TTLs supported

$2 / 1M input and $10 / 1M output started as introductory pricing through August 31, 2026, but on August 10, 2026 Anthropic made it the permanent standard rate — the scheduled increase to $3/$15 will not happen.

What's new: the intro price stuck

Released June 30, 2026, Sonnet 5 is the model Anthropic expects most developers to run by default, and it's the free-tier model in Claude.ai. It launched at $3/$15 per 1M tokens standard, with an introductory $2/$10 rate through August 31, 2026. On August 10, 2026, Anthropic reversed course: the $2/$10 rate became the permanent standard price, and the planned September 1 increase to $3/$15 was formally cancelled. Sonnet 5 now sits meaningfully cheaper than its launch pricing implied, on a permanent basis.

The Sonnet pitch is otherwise unchanged in shape: most of Opus's capability at a meaningful discount, with better latency. What moved in Sonnet 5 versus earlier Sonnets is the quality ceiling — long-horizon agent loops, tool use, and structured code edits are materially better, while the 1M context window keeps Sonnet competitive with GPT-5.4 for repo-scale work.

Capabilities

Sonnet 5 is a strong all-rounder: writing, code, vision, and structured extraction. It's the right model for production chat agents and coding assistants because answers stay composed under load, and tool use is reliable across multi-turn loops. Compared with the previous Sonnet generation, the biggest gains are in agentic execution — fewer dropped instructions and cleaner self-verification on longer tasks.

Where it still trails the Opus tier (Opus 5): the hardest novel reasoning, agentic coding, and the most demanding long-document synthesis. Where it trails GPT-5.5: outright token throughput.

Typical use cases

  • Production chat assistants and customer support agents
  • Code generation, review, and IDE integrations
  • Coding agents and long-horizon tool-use loops
  • Document Q&A and summarization (1M context for repo-scale work)
  • Vision tasks: charts, screenshots, document OCR
  • Structured extraction and form filling

Sibling and rival comparison

ModelInput / 1MOutput / 1MContext
Claude Sonnet 5$2$101M
Claude Opus 5$5$251M
Claude Haiku 4.5$1$5200K
GPT-5.4$2.50$15272K

Sonnet 5's now-permanent $2/$10 rate makes it a clear mid-tier value pick — cheaper on output than GPT-5.4, with a full 1M context window. Opus 5 costs 2.5x more for the highest-stakes work; Haiku 4.5 is half the price for narrower tasks.

← See all Anthropic / Claude plans