[go: up one dir, main page]

Released by Anthropic on June 30, 2026

Claude Sonnet 5 just launchedTry Anthropic's new model on Felo AI

Anthropic released Claude Sonnet 5 on June 30, 2026. Try the new agentic model on Felo AI today.

Claude Sonnet 5 on Felo AI
63.2%
SWE-bench Pro
Jun 30
2026 release
63.2%
SWE-bench Pro
Agentic coding benchmark reported by Anthropic
80.4%
Terminal-Bench 2.1
Stronger terminal-oriented coding performance
81.2%
OSWorld-Verified
Computer-use benchmark for agentic workflows
Jun 30
2026 release
New Sonnet model from Anthropic

Why Sonnet 5 matters

Sonnet 5 is built for work that does not stop at one chat answer. It plans, uses tools, works through code and documents, and keeps moving across multi-step tasks at a lower price point than Opus-class models.

The most agentic Sonnet yet

The most agentic Sonnet yet

Anthropic describes Sonnet 5 as its most agentic Sonnet model. It can make plans, use tools such as browsers and terminals, and run more autonomously on work that previously required larger models.

Coding that follows through

Coding that follows through

Use Sonnet 5 for repository-level coding help, debugging, test writing, refactoring, and code review. Early feedback cited stronger follow-through on messy technical contexts and multi-step software engineering work.

Better search and computer-use workflows

Better search and computer-use workflows

Sonnet 5 improves over Sonnet 4.6 on agentic search and computer-use evaluations. It gives teams more room to balance effort level, answer quality, and task cost.

Opus-like performance at a lower price

Opus-like performance at a lower price

Sonnet 5 narrows the gap with Opus 4.8 on several benchmarks while launching at lower API pricing. It is a practical default for teams that need strong execution without using the most expensive model for every task.

Model comparison

Sonnet 5 vs Sonnet 4.6 vs Opus 4.8

Sonnet 5 is not just a small Sonnet update. It moves closer to Opus 4.8 on agentic coding, computer use, reasoning, and knowledge work, while staying positioned as the everyday model for high-frequency professional workflows.

Area
Sonnet 4.6
Sonnet 5
Opus 4.8
Best for
Reliable previous-generation Sonnet workflows
Agentic coding, search, browser and terminal tasks, knowledge work
Highest-difficulty tasks where top accuracy matters most
Agentic coding
58.1% SWE-bench Pro
63.2% SWE-bench Pro
69.2% SWE-bench Pro
Terminal work
67.0% Terminal-Bench 2.1
80.4% Terminal-Bench 2.1
82.7% Terminal-Bench 2.1
Computer use
78.5% OSWorld-Verified
81.2% OSWorld-Verified
83.4% OSWorld-Verified
Pricing position
Previous Sonnet pricing baseline
Anthropic API pricing starts at $2 / $10 MTok through Aug 31, 2026, then $3 / $15 MTok standard.
$5 / $25 MTok

Built for agentic work

Use Sonnet 5 when the task has moving parts: tools, source material, code, files, constraints, partial results, and decisions that need to stay consistent across turns.

Planning

Planning

Ask Sonnet 5 to break work into steps, choose a path, track constraints, and revise when new information appears.

Tool use

Tool use

Sonnet 5 is designed for workflows that involve browsers, terminals, code execution, file context, and other tools.

Coding

Coding

Use it for debugging, reproducing issues, writing tests, reviewing pull requests, and explaining tradeoffs in existing codebases.

Research

Research

Combine Sonnet 5 with Felo Search to compare sources, extract decisions, and turn scattered information into a clear answer.

Knowledge work

Knowledge work

Use it for briefs, reports, policy analysis, legal-style research, product planning, and document-heavy professional work.

Safety

Safety

Anthropic reports lower hallucination and sycophancy rates than Sonnet 4.6, plus better resistance to malicious requests and prompt-injection hijack attempts.

What to use Sonnet 5 for on Felo

Start with work where stronger model behavior changes the outcome: code that has to run, research that must be checked, and tasks that need a model to keep going without losing the plan.

Agentic coding

Agentic coding

Ask Sonnet 5 to investigate bugs, write reproducing tests, propose fixes, review changes, and explain why an approach works.

Source-backed AI search

Source-backed AI search

Use Felo Search with Sonnet 5 to answer complex questions, compare evidence, and keep citations close to the reasoning.

Browser and terminal workflows

Browser and terminal workflows

Test prompts that involve multi-step browser or terminal-style tasks, where the model needs to choose actions and recover from partial failures.

Long research packs

Long research packs

Bring reports, product docs, meeting notes, technical references, or market research and ask for structured decisions.

Professional automation

Professional automation

Draft launch messages, update business records, summarize client context, or prepare workflows that need reliable follow-through.

Model evaluation

Model evaluation

Compare Sonnet 5 against Opus, GPT, Gemini, and other models in Felo before choosing the best model for your actual work.

Two ways to try Sonnet 5 on Felo

Felo AI Search

Open Felo AI Search with Sonnet 5 selected where available.

Open Sonnet 5 Search

Felo LLM Playground

Use the Felo LLM Playground to test Sonnet 5 with your own prompts. Compare effort, answer quality, coding behavior, and style before adding it to a workflow.

Open Playground

Frequently Asked Questions

Claude Sonnet 5 is Anthropic's newest Sonnet model, released on June 30, 2026. Anthropic positions it as the most agentic Sonnet model yet, built for coding, tool use, browser and terminal workflows, planning, and professional knowledge work.

Try Claude Sonnet 5 on Felo AI

Use Sonnet 5 for agentic coding, AI search, browser and terminal workflows, research, and long-context knowledge work.

Try Sonnet 5 on Felo

Opens Felo AI Search with Sonnet 5 selected where available