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

Anthropic dropped a new model today: Claude Sonnet 5, the latest in its Sonnet line, pitched as the most agentic Sonnet yet. It can autonomously plan, wield browsers and terminals as tools, and execute complex workflows that once demanded bigger, pricier models like Opus. The model closes a gap that had been widening, according to earlier analysis.

Sonnet models have historically been the workhorses here. Claude Sonnet 3.5, 3.6, and 3.7 were among the first to show strong coding and tool-use chops. But lately, Opus-class models have been stealing the spotlight on agentic capabilities. Sonnet 5 aims to claw that ground back: its performance is now close to Opus 4.8, but at a fraction of the cost. That pricing shift matters beyond a single release, as seen in how Kimi K3 rewired AI economics.

Performance and pricing

Sonnet 5 is live today across all plans. It is the default model on Free and Pro, and available to Max, Team, and Enterprise users. It also powers Claude Code and is accessible via the Claude API under the identifier 'claude-sonnet-5.'

Anthropic offers introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026. After that, it bumps to $3 and $15. The company says it has raised rate limits across Chat, Cowork, Claude Code, and the Claude Platform to handle the higher token consumption that comes with the model's extended reasoning at higher effort levels. The rate limit updates align with Anthropic's recent API tier simplification.

Performance benchmarks

The company shared performance numbers on two key evaluations. On the agentic search benchmark BrowseComp, Sonnet 5 (orange line) is a clear improvement over Sonnet 4.6 (gray line) and covers a much wider cost-performance spectrum than Opus 4.8 (yellow line). On the computer use evaluation OSWorld-Verified, the pattern repeats: Sonnet 5 offers substantially better cost efficiency at medium effort, and can match Opus 4.8 on some high-effort tasks.

The upshot? Users can dial the effort level up or down to balance cost and performance between Sonnet 5 and Opus 4.8.

Early adopter feedback

Early access partners have been buzzing. Testers consistently report that Sonnet 5 is more agentic than its predecessors, finishing complex tasks where earlier Sonnets would quit early, checking its own output without being asked, and doing all this at a wallet-friendly price. One tester described a session where the model investigated a bug, autonomously wrote a reproducing test, implemented the fix, then stashed the fix to confirm the bug came back without the change, all in a single pass.

"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.", Early access partner, software engineering

"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.", Early access partner, automation

"Claude Sonnet 5 gets more done with less. Same output quality, fewer steps to get there. It refuses unsafe requests cleanly and consistently.", Lovable, AI application platform

Other testers noted the model's knack for tracing failures to root causes, handling brownfield code, and shipping durable fixes rather than patching symptoms. The reliability improvement is especially relevant for researchers, some of whom already receive Claude credits via Anthropic's rare disease grants.

Safety evaluation

Anthropic's pre-deployment safety assessments found Sonnet 5 is, on the whole, an improvement over Sonnet 4.6. On agentic safety, it is better at refusing malicious requests and resisting hijack attempts during prompt injection attacks. It also shows lower rates of hallucination and sycophancy.

On the company's automated behavioral audit, which tests a wide range of misaligned behaviors, cooperation with misuse, deception, and the like, Sonnet 5 scored lower (safer) overall. But it did show somewhat higher rates of misaligned behavior compared to the more capable Opus 4.8 and Claude Mythos Preview.

Anthropic did not deliberately train Sonnet 5 on cybersecurity tasks. It can handle some routine, non-harmful cyber tasks, but on evaluations probing potentially dangerous cyber skills, such as developing software exploits, it performs substantially worse than models like Opus 4.8 and Mythos 5. In one test looking at the ability to develop exploits for Firefox browser vulnerabilities, Sonnet 5 was never able to produce a full working exploit, though it did show a slightly higher rate of partial success than Sonnet 4.6. Anthropic attributes this to improvements in general intelligence rather than specific training. The broader strategy for handling dangerous capabilities is illustrated by Mythos 5's strict access controls.

Because Sonnet 5 is somewhat stronger than its predecessor on these tasks, Anthropic has launched it with cyber safeguards enabled by default. These safeguards detect and block dangerous cyber usage in real time, and are the same ones present in Claude Opus 4.7 and 4.8.

Availability and changelog

The model is available immediately through Anthropic's native platform, AWS, Microsoft Foundry (hosted on Azure and Anthropic), and coming soon on Google Vertex. Organizations already enrolled in the Cyber Verification Program get automatic access on Sonnet 5, with no need to reapply. The enterprise push echoes the approach used with TCS's 50,000-employee Claude deployment.

A footnote in the announcement notes that while Sonnet 5 is an upgrade to Sonnet 4.6, it uses an updated tokenizer that may result in roughly 1.0 to 1.35 times more tokens depending on content type. The introductory pricing, the company says, is designed to make the transition roughly cost-neutral.