Claude Sonnet 5 Online | Try Anthropic's New AI
Get Started Free

Claude Sonnet 5

Current version: 5.0 updated

Anthropic's workhorse for agents and coding

Category
Language
Modality
Text · Tool use
Context
1M
Released
Jun 2026
Run this model Estimate cost
Update

What's new in Claude Sonnet 5 5.0

  • Agentic coding around 63 percent vs 58 for 4.6 — closing in on Opus-level results
  • Finishes complex multi-step tasks and verifies its own output
  • Stronger with browsers, terminals and external tools
  • Lower hallucination rates and better prompt injection resistance
  • 1M token context as standard
Strengths

What it's the best tool for

  • 1M token context — whole projects in a single request
  • Near-Opus agentic coding at a much lower price
  • Verifies and fixes its own output unprompted
  • Confident with browsers, terminals and tool calls
  • Lower hallucinations and stronger prompt injection resistance
Limitations

When to reach for something else

  • No native image or video generation
  • Soft content filter on borderline topics
  • Knowledge limited to training cutoff, no built-in web access
  • Updated tokenizer uses slightly more tokens for the same text
Sample output

How Claude Sonnet 5 responds

Prompt
Act as a senior engineer. Take this Python repo, find why the integration tests fail, fix the code and explain exactly what you changed.
Claude Sonnet 5
Root cause found: the DB fixture is created before migrations run — 7 tests fail on a missing column. Fixed the init order in conftest.py and added the missing migration. Re-ran the suite: 214 passed, 0 failed. Bonus: two tests were flaky due to timezone drift — pinned UTC in settings.
Where teams use it

Four scenarios where it pays for itself

01
Agents and automation
Planning, browser, terminal and tools in one loop
02
Production-grade coding
Refactors and migrations on 1M token context
03
Web form workflows
Multi-step scenarios across multiple tabs
04
Long research
Large PDFs, data rooms and reports in one request
About model

More about Claude Sonnet 5

Claude Sonnet 5 Online — Anthropic's New Workhorse Model

Claude Sonnet 5 is Anthropic's latest model, released June 30, 2026. It succeeds Sonnet 4.6 in the workhorse tier with a major leap in agentic capability: the model plans ahead, drives browsers and terminals, calls tools and finishes multi-step tasks where earlier Sonnet models stopped short. Try it on NetRoom online, no VPN required.

What's new vs Sonnet 4.6

The headline is autonomy. On agentic coding benchmarks Sonnet 5 scores around 63 percent versus 58 for Sonnet 4.6, closing most of the gap to Opus while staying much cheaper. It verifies its own output without being asked — spotting a failing test, fixing the code and re-running the suite. Hallucination and sycophancy rates are down, and resistance to prompt injection is up, which matters for agents browsing the open web.

1M token context

Sonnet 5 handles up to 1M tokens of context — an entire monorepo, a data room or a stack of PDFs in a single request. For large refactors and long research sessions that means no chunking and no lost continuity between parts of the input.

When to pick it

Choose Sonnet 5 for end-to-end agent work: repo-wide reviews, multi-file migrations, multi-tab form filling, report generation over large document sets. For quick chat and simple copy, cheaper models will do; for maximum quality regardless of cost, look at Opus. Sonnet 5 is the sweet spot between the two — which is exactly why Anthropic made it the default model in its own products.

How to start

Open NetRoom, pick Claude Sonnet 5 from the catalog and work right in your browser. No subscription and no foreign accounts needed: pay per token used, keep your chat history and combine the model with image generation and document tools in one interface.

Versions

Version history of Claude Sonnet 5

Version Date What changed
5.0 current
  • Agentic coding around 63 percent vs 58 for 4.6 — closing in on Opus-level results
  • Finishes complex multi-step tasks and verifies its own output
  • Stronger with browsers, terminals and external tools
  • Lower hallucination rates and better prompt injection resistance
  • 1M token context as standard
4.6
  • Best-in-class computer use and multi-tab agents
  • Up to 1M token context in beta
  • Strong agentic coding and large-project refactors

Try Claude Sonnet 5
right now

Free access to basic models. No card, no obligations.