Skip to main content
Release date: July 1, 2026 LLMEasy has updated the Claude Sonnet 5 model entry. Check the model plaza for the currently available model ID and live pricing, then configure it through the Claude Code or external-tool setup guide that matches your client. Claude Sonnet 5 is the next generation of the Sonnet family. Anthropic positions it as an upgrade from Claude Sonnet 4.6, with the largest improvements in coding, agentic tasks, and professional work.

​
Highlights

  • Better fit for coding and agent work: Anthropic says Sonnet 5’s largest gains over Sonnet 4.6 are in coding and agentic tasks
  • Larger context window: Sonnet 5 supports a 1M tokens context window by default and up to 128K tokens of output
  • Adaptive thinking is on by default: Requests without a thinking field run with adaptive thinking
  • Low migration effort: If you already use Sonnet 4.6, most requests can move by changing the model ID to claude-sonnet-5
  • Clearer safety behavior: For high-risk cybersecurity content, the model may return a refusal with stop_reason: "refusal" instead of an HTTP error

​
Migration notes

If you are moving from Sonnet 4.6 to Sonnet 5, review these settings first:
  • Revisit max_tokens: The thinking trace and final answer share the same output limit, so tasks close to the limit may need more room
  • Remove manual thinking budgets: thinking: { "type": "enabled", "budget_tokens": ... } is no longer supported on Sonnet 5; use adaptive thinking instead
  • Do not set non-default sampling parameters: Non-default temperature, top_p, or top_k can return a 400 error
  • Recount tokens: Sonnet 5 uses a new tokenizer, so the same text can produce more tokens than on Sonnet 4.6

​
Setup guidance

If you use Claude Code, start with the Claude Code setup guide. Different Key groups use different protocols and model configuration patterns:
  • Claude group: use https://www.llmeasy.ru
  • GPT group or external tools: follow the relevant tool guide with https://www.llmeasy.ru/v1
Use the model name currently shown in the model plaza. Do not hardcode a model name that is not shown there.
If you run long-context coding tasks, automatic fixes, code review, or multi-step agent workflows, test claude-sonnet-5 first. If your workload is cost-sensitive, estimate cost from the model plaza pricing before switching.

​
References