Claude Sonnet 5 launched today as the most agentic Sonnet model to date. It is designed to make plans, use tools such as browsers and terminals, and run autonomously on multi‑step tasks at a level that previously required larger, more expensive models.
Performance and positioning
Sonnet‑class milestones earlier included Sonnet 3.5, 3.6 and 3.7, which demonstrated notable coding and tool‑use abilities; more recently, the clearest gains in agentic capabilities have come from Opus‑class models. Sonnet 5 narrows that gap: its performance approaches Opus 4.8 while offering lower pricing.
Compared with Sonnet 4.6, Sonnet 5 represents a substantial improvement on agentic aspects such as reasoning, tool use, coding and knowledge work. The announcement materials present comparative results across several evaluations versus Sonnet 4.6 and Opus 4.8; a broader set of measurements is available in the Claude Sonnet 5 System Card.
Safety and cybersecurity
Pre‑deployment safety assessments found Sonnet 5 shows an overall lower rate of undesirable behaviors than Sonnet 4.6. It is better at refusing malicious requests and resisting hijack attempts in prompt injection attacks, and shows lower rates of hallucination and sycophancy.
However, on an automated behavioral audit that tests a wide range of misaligned behaviors, Sonnet 5 scored higher (i.e., worse) than Opus 4.8 and Claude Mythos Preview on some measures. Sonnet 5 was not deliberately trained on cybersecurity tasks: it can perform some routine, non‑harmful cyber actions, but on evaluations of potentially dangerous cyber skills (for example, developing software exploits) it performs substantially worse than Opus 4.8 and Mythos 5.
In a test developed in collaboration with Mozilla, models were evaluated on developing exploits for vulnerabilities in Firefox 147 (these vulnerabilities were later patched in Firefox 148). Neither Sonnet model developed a working exploit (both scored 0.0% for full exploit success); Sonnet 5 did show a slightly higher partial success rate than Sonnet 4.6. This change is likely attributable to general intelligence improvements rather than targeted training.
Because Sonnet 5 is somewhat stronger on these tasks than its predecessor, it launches with cyber safeguards enabled by default. These real‑time detection and blocking safeguards are the same as those used for Claude Opus 4.7 and 4.8, though they are less strict than the broader safeguards deployed with Fable 5.
A full account of Sonnet 5’s safety and capability evaluations is reported in the Claude Sonnet 5 System Card.
Availability and pricing
Claude Sonnet 5 is available everywhere today. It is the default model for Free and Pro plans and is accessible to Max, Team and Enterprise users. It is also available in Claude Code and on the Claude Platform.
Introductory pricing runs through August 31, 2026: $2 per million input tokens and $10 per million output tokens. After that date, standard pricing will be $3 per million input tokens and $15 per million output tokens.
The cost‑performance charts in the announcement show Sonnet 5 priced at $3/MTok input and $15/MTok output; with the introductory pricing the effective cost is even lower. For comparison, Opus 4.8 is shown at $5/MTok input and $25/MTok output.
Developer access and limits
Developers can access claude‑sonnet‑5 via the Claude API. Rate limits were increased for Sonnet and Haiku on April 26, 2026, and the native Claude Platform simplified limits into three tiers (Start, Build, Scale) to support higher token usage at elevated effort levels. The Sonnet 5 model uses an updated tokenizer (as introduced previously with Opus 4.7) that improves model processing but can map the same input to roughly 1.0–1.35× more tokens depending on content type; the introductory pricing is intended to make the transition roughly cost‑neutral.
Early partner feedback
Early access partners consistently reported that Sonnet 5 is significantly more agentic than earlier Sonnet models: it completes complex tasks that previously stalled, checks its own outputs without explicit instruction, and does so at an attractive price point. Reported use cases include sustained coding, tool use, debugging in messy technical contexts, legal research, live data exploration and insurance workflow automation. Several named testers commented on specific benefits, including Zimu Li, Daniel Shepard, Fabian Hedin, Yusuke Kaji, Neel Chotai, Sualeh Asif, Dominic Elm, Mauricio Wulfovich, Ryadh Dahimene and Eric He.
Summary
Claude Sonnet 5 is the newest Sonnet‑class model, bringing stronger agentic capabilities closer to Opus‑level performance at lower cost. Safety evaluations indicate overall improvements compared with Sonnet 4.6, while cyber capabilities remain weaker than Opus 4.8, so default cyber safeguards are in place. The model is available now across plans and platforms with introductory pricing through August 31, 2026.
