Anthropic Launched Sonnet 5. It's Almost at Opus 4.8 Level — For Half the Price.

Anthropic didn't announce it with fanfare. Today, June 30, 2026, it simply rolled out Claude Sonnet 5 across every plan. And it changed the rules of the game in the segment that matters most to anyone building with AI: cost per task.

The promise is direct. Sonnet 5 closes in on the performance of Opus 4.8, the house's flagship model, at a much lower price. For years, the biggest leaps in agentic capability came from Opus models, more expensive and heavier. Sonnet 5 closes that gap and puts frontier capability within reach of normal budgets.

What changes from Sonnet 4.6

This is a substantial upgrade over its predecessor on everything that matters for real work: reasoning, tool use, coding and knowledge work. Anthropic describes it as the most agentic Sonnet yet, capable of making plans, using tools like browsers and terminals, and running autonomously at a level that a few months ago required much bigger, more expensive models.

Early-access partners were consistent in their description: Sonnet 5 finishes complex tasks where earlier Sonnets stalled halfway, and checks its own output without being asked. In one account, the model investigated a bug, wrote a test that reproduced it, implemented the fix, and confirmed the result, all in a single pass.

Coding and agents: where it shines

Coding and autonomous agents are where Sonnet 5 shows its clearest gains. On Anthropic's cost-performance curves in evaluations like BrowseComp, for agentic search, and OSWorld-Verified, for computer use, Sonnet 5 is a strict improvement over Sonnet 4.6. Opus 4.8 remains the choice for maximum precision, but now both share the same quality range, with Sonnet 5 offering a much cheaper entry point.

The technical community reacted strongly. On social media, developers nicknamed the model Fennec and highlighted its speed inside Claude Code, with several citing an 82.1% on SWE-bench Verified milestone as a benchmark for its ability to solve real engineering problems. The enthusiasm centers on one point: the same output quality, with fewer steps to get there.

Price: the real move

Here's the punch. Sonnet 5 launches with an introductory price of $2 per million input tokens and $10 per million output tokens, valid until August 31, 2026. After that, it moves to $3 and $15 respectively. For comparison, Opus 4.8 costs $5 and $25.

Translation: flagship-comparable capability, at a fraction of the cost. For high-volume workloads, where each task multiplied by thousands weighs on the bill, this is the difference between viable and prohibitive. One technical note worth keeping: Sonnet 5 uses an updated tokenizer, which can slightly increase the token count for the same text, but the introductory price was calibrated so the transition stays roughly cost-neutral.

Already everywhere

There's no waitlist. As of today, Sonnet 5 is the default model on the Free and Pro plans, and it's available for Max, Team and Enterprise. Developers can use it via the Claude Platform with the identifier claude-sonnet-5, and it's integrated into Claude Code and Cowork. Anthropic also raised usage limits to accommodate the higher effort levels.

On safety, the company says Sonnet 5 shows a lower overall rate of undesired behaviors than Sonnet 4.6, with fewer hallucinations and less sycophancy. It launched with cybersecurity safeguards active by default, though its capabilities in that domain are considerably below the Opus models.

The takeaway for anyone building AI products is clear: what needed the most expensive model until yesterday can now run on a model at half the price. And the discount clock is ticking.