Anthropic Just Made Its Flagship AI Harder to Recommend

Anthropic just priced its cheaper model at $2 per million input tokens and $10 per million output tokens, and claimed it lands close to the performance of Opus 4.8, a model that costs $5 and $25 for the same volumes. On June 30, 2026, the company released Claude Sonnet 5 and described it as the most agentic Sonnet it has ever built. For a lot of teams, the interesting question is no longer whether Sonnet is good enough, but whether the premium flagship still earns its price.

A Sonnet That Closes the Gap on Opus

Anthropic frames the release around a shift in where agentic gains have been coming from. In its telling, the agentic AI era began with Sonnet-class models, Claude Sonnet 3.5, 3.6, and 3.7 were the first to show strong coding and tool use, but the clearest recent gains had moved to the more expensive Opus tier. Sonnet 5 is positioned to narrow that gap: the company says its performance is close to that of Opus 4.8 but at lower prices, and calls it a substantial improvement over its predecessor, Sonnet 4.6, on reasoning, tool use, coding, and knowledge work.

Built to run on its own

The headline capability is autonomy. Anthropic says Sonnet 5 can make plans, use tools like browsers and terminals, and run autonomously at a level that just a few months ago required larger and more expensive models. Early access partners echoed the theme. One tester, Neel Chotai, said he asked Sonnet 5 to investigate a bug, and 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. Another, Daniel Shepard, described handing it a two-part job, updating Salesforce account tiers and sending a launch announcement to enterprise contacts, that it finished end to end where previous models used to stall halfway.

The Pricing Is the Real Announcement

Claude Sonnet 5 is available everywhere from launch day. It is the default model for Free and Pro plans and is available to Max, Team, and Enterprise users, as well as in Claude Code and on the Claude Platform, where developers call it with the identifier claude-sonnet-5. On the API it launched with introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026, after which it moves to $3 per million input and $15 per million output. For reference, Opus 4.8 is priced at $5 per million input and $25 per million output, roughly double.

The tokenizer caveat worth knowing

There is a detail buried in the footnotes that developers should not miss. Sonnet 5 uses an updated tokenizer that changes how the model processes text, which improves performance but can map the same input to more tokens, roughly 1.0 to 1.35 times depending on content type. Anthropic says it set the introductory pricing so the transition from Sonnet 4.6 is roughly cost-neutral. In practice that means the sticker price and your actual bill can diverge, so teams migrating should measure real token consumption rather than assume a flat discount.

Safety Shipped On by Default

Anthropic’s pre-deployment evaluations found Sonnet 5 to be safer overall than Sonnet 4.6. The company reports it is better at refusing malicious requests and resisting hijack attempts in prompt-injection attacks, and shows lower rates of hallucination and sycophancy. On an automated behavioral audit testing a wide range of misaligned behaviors, Sonnet 5 scored lower, meaning safer, than 4.6, though still higher than the more capable Opus 4.8 and Claude Mythos Preview.

On cybersecurity, Anthropic says it did not deliberately train Sonnet 5 on cyber tasks. In an evaluation developed with Mozilla that tested whether models could build exploits for vulnerabilities in Firefox 147, neither Sonnet 5 nor Sonnet 4.6 could develop a working exploit, both scored 0.0 percent, though Sonnet 5 showed a slightly higher partial-success rate. Because the model is somewhat stronger than its predecessor, Anthropic launched it with real-time cyber safeguards enabled by default, the same set used in Claude Opus 4.7 and 4.8. All the tested Firefox vulnerabilities have since been patched in Firefox 148.

Related

• Claude Science, Anthropic’s AI Workbench for Researchers
• Saudi Aramco Just Bet $800M Against Closed AI
• Ford Handed Quality Control to AI. Now It’s Reversing Course.
• Qualcomm’s Real Target Wasn’t the Chip
• Why Wall Street Thinks Micron Is the Next Nvidia

Sources

1. Anthropic — Introducing Claude Sonnet 5 (Jun 30, 2026)
2. Anthropic Newsroom
3. TechCrunch — Artificial Intelligence