Claude Sonnet 5: Performance, Pricing and What’s New
TL;DR: Claude Sonnet 5 launched on June 30, 2026 across all Anthropic plans. The model reaches performance close to Claude Opus 4.8 at a lower cost: $2/MTok input and $10/MTok output through August 31, 2026, then $3 and $15. Built for autonomous agents, it significantly improves reasoning, tool use and coding over Sonnet 4.6.
Source:
Introducing Claude Sonnet 5 – Anthropic
· Published June 30, 2026
Anthropic launched Claude Sonnet 5 on June 30, 2026 with a clear positioning: the most agentic Sonnet model to date, capable of autonomous planning and tool use (browsers, terminals) at performance levels approaching the Opus range, at a lower price point. This is a meaningful upgrade from Sonnet 4.6, with direct implications for the cost/performance ratio of teams deploying agents in production.
The model is deployed across Claude Platform, Claude Code, and native Claude applications. If you are currently using Sonnet 4.6, migration to Sonnet 5 is straightforward, with one point to watch: the updated tokenizer uses 1.0 to 1.35 times more tokens depending on content type, which affects actual cost per request.
What improvements does Sonnet 5 bring over Sonnet 4.6?
Anthropic describes substantial gains across four areas: reasoning, tool use, coding, and knowledge work. On internal evaluations, Sonnet 5 completes complex multi-step tasks that Sonnet 4.6 would not finish.
A concrete example documented by Anthropic: a Rust engineer reported that the model “wrote a reproducing test, implemented the fix, then stashed it to confirm the bug came back without the change. All in a single pass.” This type of reasoning-verification-validation loop is precisely what distinguishes Sonnet 5 from its predecessor. On automated coding workflows, Sonnet 5 checks its own output without being explicitly asked, reducing back-and-forth cycles.
In a business workflow context, a user reported a Salesforce update and enterprise announcement task “finished end to end, that used to stall halfway.” ClickHouse noted that the model “reasons in tighter steps and gets users to answers noticeably faster” in their analytics use cases. These testimonials point to a real improvement in agentic reasoning, not just on synthetic benchmarks.
Can Sonnet 5 replace Opus 4.8?
On certain tasks, yes. According to Anthropic, Sonnet 5 can match Opus 4.8 capability levels at higher effort settings. This broadens the range of use cases a Sonnet-tier model can handle without upgrading to Opus.
The cost difference remains significant: Sonnet 5 at $3/MTok input versus Opus 4.8 at $15/MTok input (standard pricing). For high-volume workflows or multi-agent systems with many calls, the choice between the two becomes an economic decision as much as a technical one. Our article on Claude Opus 4.8 details the specific use cases where the Opus range holds the advantage.
One point worth noting: Sonnet 5’s cybersecurity capabilities are intentionally reduced compared to Opus 4.8, with safety guardrails enabled by default. During Firefox vulnerability tests, Sonnet 5 produced zero successful exploits. For teams working on offensive security or red-teaming, Opus 4.8 remains the reference. For standard agentic workflows, Sonnet 5 covers the scope at a much lower cost.
What is the pricing for Claude Sonnet 5?
Two pricing tiers are in effect. Through August 31, 2026, the introductory price is $2/MTok input and $10/MTok output. Starting September 1, 2026, standard pricing applies: $3/MTok input and $15/MTok output.
| Period | Input | Output |
|---|---|---|
| Through August 31, 2026 (intro) | $2 / MTok | $10 / MTok |
| From September 1, 2026 | $3 / MTok | $15 / MTok |
The model is available on all Anthropic plans: Free, Pro, Max, Team, and Enterprise. Rate limits have been increased across all platforms to accommodate higher-effort usage.
For cost calculations, the new tokenizer coefficient (1.0 to 1.35x) applies to token volume. Content that consumed 1,000 tokens with Sonnet 4.6 can consume up to 1,350 with Sonnet 5. At equivalent content volume, actual cost will therefore be slightly higher than the per-token price alone suggests. Benchmarking on your actual requests is recommended before any cost projection at scale.
Which use cases is Sonnet 5 designed for?
The agentic angle is central to the announcement. Sonnet 5 is documented as performing well across several task categories: multi-step business workflow automation (CRM, API integrations, reporting), data analysis with iterative reasoning, coding with autonomous verification, legal research and document analysis (described as sitting on the “Pareto frontier for plaintiff-law tasks” by a legal-tech user), and insurance workflows via computer-use (submission intake, FNOL, loss runs).
For teams orchestrating agents with MCP tools or frameworks like LangGraph and CrewAI, Sonnet 5 should perform notably better than Sonnet 4.6 on long tool chains. If you are deploying agents in production and runtime security is a priority, our guide on AI agent security covers layered architecture, sandboxing, and the autonomy levels to define before any deployment.
How does Anthropic approach security in Sonnet 5?
The announcement highlights three security outcomes: a lower rate of misaligned behaviors than Sonnet 4.6 according to internal evaluations, zero successful exploits during Firefox vulnerability tests, and cybersecurity capabilities intentionally reduced compared to Opus 4.8 with guardrails enabled by default.
This positioning is consistent with Anthropic’s strategy since Opus 4.8: Sonnet models are calibrated for broad-public agentic use with a controlled risk profile, while Opus models retain more extensive capabilities for specialized contexts. For teams integrating Sonnet 5 into agentic pipelines with access to sensitive tools (filesystem, financial APIs, infrastructure), runtime security policies remain your responsibility. Our article on the MCP protocol explains how to structure tool access for your agents.
Frequently asked questions
Is Claude Sonnet 5 available on the free plan?
Yes. Anthropic deployed Sonnet 5 across all plans including Free. Rate limits vary by plan, and limits have been increased on all platforms to accommodate the higher token volume of the updated tokenizer.
What is the concrete difference between Sonnet 5 and Opus 4.8?
Sonnet 5 approaches Opus 4.8 performance on agentic tasks at roughly one-fifth the cost at standard pricing ($3 versus $15/MTok input). Opus 4.8 retains an advantage on the most demanding tasks and cybersecurity capabilities. For most automation and coding workflows, Sonnet 5 covers the scope at a lower cost.
Do I need to recalculate costs when migrating from Sonnet 4.6 to Sonnet 5?
Yes. The new tokenizer uses 1.0 to 1.35 times more tokens than Sonnet 4.6 for equivalent content. The per-token price is comparable, but request volume increases depending on content type. Benchmarking on your actual requests is recommended before any cost projection at scale.
Sonnet 5 repositions the Sonnet range as a viable option for serious agentic workflows, without systematically upgrading to Opus. For teams orchestrating agents with multiple tools, the progress on multi-step reasoning and autonomous verification concretely changes what you can delegate without constant human oversight. The introductory pricing window through August 31 is worth factoring in if you are planning a migration from Sonnet 4.6.
