Claude Sonnet 5 just made the Opus price gap harder to justify

Anthropic has released Claude Sonnet 5, its newest mid-tier model designed to dramatically improve agentic capabilities, planning, tool use, and autonomous operation, without the price tag of its Opus-class siblings. Available today across Free, Pro, Max, Team, and Enterprise plans, as well as in Claude Code and via the Claude API, Sonnet 5 is marketed as narrowing the performance gap with Opus 4.8 while offering substantial cost savings.

According to Anthropic, the model marks a significant leap over its predecessor, Sonnet 4.6, on key metrics including reasoning, tool use, coding, and knowledge work. The company's internal evaluations show Sonnet 5 outperforming Sonnet 4.6 across a wide range of benchmarks, including agentic search and computer use evaluations. In some cases, higher-effort Sonnet 5 performance matches that of Opus 4.8.

Pricing and availability

Sonnet 5 launches with introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026. After that, standard pricing kicks in at $3 per million input tokens and $15 per million output tokens. Anthropic notes that Sonnet 5 uses an updated tokenizer, similar to the change introduced with Opus 4.7, which may increase token counts by roughly 1.0, 1.35x depending on content. The introductory pricing is designed to make the transition roughly cost-neutral for users.

The model is the default for Free and Pro users and is available to Max, Team, and Enterprise subscribers. Anthropic has also raised rate limits across Chat, Cowork, Claude Code, and the Claude Platform to handle higher token usage at elevated effort levels. Users can adjust the effort parameter to balance cost and performance.

Agentic performance: Testimonials from early partners

Feedback from Anthropic's early access partners has been consistently positive. Developers and organizations described a model that finishes complex tasks where previous Sonnet models would stop short, self-checks its output without prompting, and performs autonomously at an attractive price point.

Joe Spisak, Product Director at Meta, said: Claude Sonnet 5 gives our agents a strong execution layer for multi-step software engineering work. It handles sustained coding, tool use, and debugging well across messy technical contexts, and has been especially useful for workflows where follow-through and technical grounding matter.

CoSailor described a real-world test: We handed Claude Sonnet 5 a two-part job, update Salesforce account tiers, send a launch announcement to enterprise contacts, and it finished end to end. That used to stall halfway.

Lovable founder Henrik Torstensson praised the model's safety behavior: Claude Sonnet 5 gets more done with less. Same output quality, fewer steps to get there. It refuses unsafe requests cleanly and consistently, too.

Stripe's David Byrne noted the model's autonomous debugging prowess: 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.

Notion Labs' Simon Last highlighted its reliability in multi-step code changes: With Claude Sonnet 5, agents stay on plan, follow our conventions, and ship clean multi-step changes, all at an efficient cost.

Cognition AI's Scott Wu said: Claude Sonnet 5 is at its best on brownfield code, race conditions, hidden tests, the parts nobody wants to touch. It traces a failure to its actual root cause and ships a durable fix instead of patching the symptom.

Eve Legal's Peter Wang noted gains in legal research: Claude Sonnet 5 sits on the Pareto frontier for Eve's plaintiff-law tasks. We see the clearest gains in legal research and analysis, at a price-to-performance ratio that made the choice to migrate easy.

ClickHouse's Nick Ursa emphasized speed: Claude Sonnet 5 reasons in tighter steps and gets our users to answers noticeably faster.

Pace CEO Jack Terrell observed the model's reliability in production: Our computer-use agents run insurance workflows ... Claude Sonnet 5 consistently takes the right action and does it quickly, which is what real insurance work demands.

Safety evaluation and cyber safeguards

Anthropic's pre-deployment safety assessments found Sonnet 5 to be safer overall than Sonnet 4.6, with lower rates of hallucination, sycophancy, and misaligned behavior. The model also showed improvement in refusing malicious requests and resisting prompt injection attacks. However, the automated behavioral audit, which tests a wide range of undesirable behaviors, found Sonnet 5 to have slightly higher rates of misaligned behavior than the more capable Opus 4.8 and the unreleased Claude Mythos Preview.

In cybersecurity evaluations, Sonnet 5 demonstrated substantially poorer performance than Opus 4.8 and Mythos 5 on tasks such as developing software exploits. For example, in a test involving Firefox browser vulnerabilities, neither Sonnet 4.6 nor Sonnet 5 ever produced a full working exploit, both scored 0.0%, though Sonnet 5 showed a marginally higher partial success rate. Anthropic attributes this to general intelligence improvements rather than specific training in cyber capabilities.

Because Sonnet 5's partial cyber success rate is slightly higher than its predecessor's, Anthropic has launched the model with cyber safeguards enabled by default. These safeguards, detecting and blocking dangerous cyber usage in real time, are the same as those used with Opus 4.7 and 4.8. They are less strict than the safeguards deployed with Fable 5, which block a much wider range of cybersecurity tasks. Anthropic recommends Opus 4.8 for cybersecurity work that requires reduced guardrails.

Changelog and methodology note

Anthropic issued a post-launch correction: the original version of the blog post included a cost-performance chart for the BrowseComp evaluation that used a simpler methodology, underestimating Sonnet 5's performance. The chart has been updated to match the standard methodology described in the system card, which uses a 10 million token budget with compaction and programmatic tool calling.

Additionally, Sonnet 4.6 scores on Humanity's Last Exam and OSWorld-Verified have been updated to reflect changes in the grader model and evaluation methodology respectively.

The Claude Sonnet 5 System Card, which reports full evaluation details and safety assessments, is available on Anthropic's website.