Summary: Anthropic shipped Claude Sonnet 5 on June 30, 2026 as its most agentic Sonnet model, priced at $2 per million input tokens and $10 per million output tokens. Effort levels widen its cost-performance curve, and partners reported better follow-through on coding, legal, insurance, and data work. It launched with cyber safeguards because it was not trained for advanced cyber offense and scored zero on a Firefox exploit-development evaluation.
What Shifted: A New Cost-Performance Frontier
Anthropic shipped Claude Sonnet 5 on June 30, 2026, billing it as the most agentic Sonnet model to date. Plenty of developers trace the start of their agentic AI work back to Sonnet-class models: Claude Sonnet 3.5, 3.6, and 3.7 were the first in the line to show off striking coding and tool-use ability. Since then, though, most of the movement in agentic capability has happened in Opus-class models. Sonnet 5 closes much of that distance — it performs close to Opus 4.8 while costing less.
The price is $2 per million input tokens and $10 per million output tokens, a rate that became permanent on August 10, 2026 (an earlier plan to raise it to $3/$15 in September was scrapped). Set against its predecessor, Sonnet 4.6, that marks a substantial improvement on the dimensions that matter most for agentic work: reasoning, tool use, coding, and knowledge work.
How It Works: Effort Levels and Benchmarks
Anthropic’s comparison put Sonnet 5, Sonnet 4.6, and Opus 4.8 head to head on the agentic search evaluation BrowseComp and the computer use evaluation OSWorld-Verified, with each run at different “effort” levels. That effort parameter is the dial for how much compute the model spends, trading cost against performance.
On these benchmarks, Sonnet 5 (orange line) is a strict improvement over Sonnet 4.6 (gray line), and the range of cost-performance options it covers is far wider than that of Opus 4.8 (yellow line). Cost efficiency improves substantially at medium effort; push effort higher and its performance can match Opus 4.8 on some tasks. The upshot: you can dial in the balance of cost and performance that fits your specific use case.
Early access partners came back saying Sonnet 5 is much more agentic than the models before it. Testers described how it finishes complex tasks where previous Sonnet models would stop short, how it checks its own output without being asked, and how it does all this at an attractive price point. For example, one tester said: “I asked Claude Sonnet 5 to investigate a bug. Unprompted, it wrote a reproducing test, implemented the fix, then stashed it to confirm the bug came back without the change. All in a single pass.”
Practical Use Cases: From Software Engineering to Insurance
Partner feedback highlights Sonnet 5’s follow-through and cost efficiency across domains:
- Software engineering: One partner observed that it “handles sustained coding, tool use, and debugging well across messy technical contexts” — a particular advantage for workflows where follow-through and technical grounding are what decide the outcome. A second partner reported that it “carried each one through to a tested, verified result on its own,” which leaves engineers free to concentrate on judgment and final sign-off.
- Legal research: Eve, a legal AI platform, described Sonnet 5 as a model that “sits on the Pareto frontier” for plaintiff-law tasks; the clearest gains showed up in legal research and analysis, at a price-to-performance ratio that made migration easy.
- Insurance automation: Pace, which runs computer-use agents for insurance workflows, said Sonnet 5 “consistently takes the right action and does it quickly” — exactly what real insurance work demands.
- Data exploration: ClickHouse’s agents lean on Sonnet 5 to explore live data and produce insights on the fly; as they put it, it “reasons in tighter steps and gets our users to answers noticeably faster.”
Taken together, these cases show Sonnet 5 taking on multi-step, end-to-end work that used to call for pricier models, which puts agentic automation within easier reach.
Limitations and Trade-offs
Anthropic’s pre-deployment safety evaluations found Sonnet 5 overall safer than Sonnet 4.6. It refuses malicious requests more reliably and resists prompt injection attacks better, and its rates of hallucination and sycophancy are lower. In their automated behavioral audit, Sonnet 5 scored lower — that is, safer — overall than Sonnet 4.6, while its rate of misaligned behavior ran somewhat higher than Opus 4.8 and Claude Mythos Preview.
One thing worth noting: Anthropic did not deliberately train Sonnet 5 on cybersecurity tasks. On evaluations that test dangerous cyber skills, such as developing software exploits, it performed substantially worse than Opus 4.8 and Mythos 5. In a Firefox exploit development evaluation, Sonnet 5 never successfully developed a full working exploit (0.0% success), though it had a slightly higher partial success rate than Sonnet 4.6. On that basis, Anthropic launched Sonnet 5 with cyber safeguards enabled by default, the same as those in Opus 4.7 and 4.8, but less strict than those for Fable 5.
What this means for product builders: for general agentic tasks, Sonnet 5 is a strong choice. But if your needs run to advanced cybersecurity capabilities (e.g., penetration testing), you may still want Opus-class models. And since its misaligned behavior rates, though below Sonnet 4.6, sit above Opus 4.8, high-risk applications warrant careful evaluation.
Concrete Takeaway: Test Effort Levels on Your Workload
Claude Sonnet 5 has rolled out across all plans: Free and Pro get it as the default, and Max, Team, and Enterprise users can use it too. On the Claude API, developers reach it by calling the model name claude-sonnet-5. Anthropic has also increased rate limits across Chat, Cowork, Claude Code, and the Claude Platform, so that the heavier token usage of higher effort levels has room to run.
If Sonnet 5 is on your evaluation list, the important step is testing different effort levels against the tasks you actually have. Its cost-performance curve is wider than Opus 4.8’s, which leaves room to find a sweet spot that balances cost and quality. Across many workloads, Sonnet 5 may deliver near-Opus performance at a fraction of the cost, making it a practical choice for scaling agentic AI in production.
For full details, refer to the Claude Sonnet 5 System Card, which reports a broader set of evaluations and safety results.