Anthropic's Claude Sonnet 5 Closes In on Opus 4.8 at a Fraction of the Value – Decrypt

Anthropic launched Claude Sonnet 5 on Tuesday, calling it “essentially the most agentic Sonnet mannequin but.” It is the default mannequin for Free and Professional customers, stay on Max, Staff, and Enterprise plans, in Claude Code, and thru the API . In contrast to previous Sonnet launches, this one is constructed to take a seat subsequent to the earlier Opus as an alternative of trailing a tier behind it.

In its launch submit, the corporate says Sonnet 5’s efficiency is “near that of Opus 4.8, however at decrease costs.” Builders can slide an effort dial between the 2 fashions or select totally different ranges on the internet app to commerce value for accuracy on the identical activity, masking floor that used to require Opus charges.

On SWE-bench Professional—a coding benchmark pulling issues from actively maintained repositories with multi-file adjustments, scored as p.c solved—Sonnet 5 hit 63.2% towards Sonnet 4.6’s 58.1%.

On GDPval-AA v2, an Synthetic Evaluation benchmark that scores real-world skilled duties throughout 44 jobs through blind pairwise Elo rankings, it landed at 1,618, a statistical tie with Opus 4.8’s 1,616. The variations between Sonnet 5 and Opus 4.8 on Humanity’s Final Examination are principally negligible: 57.4% vs 57.9%.

Sonnet 5 additionally ships with an up to date tokenizer—the system that breaks textual content into the models a mannequin payments for—and it is hungrier, turning the identical enter right into a activity that consumes extra tokens. “Sonnet 5 is an improve to Sonnet 4.6, however it makes use of an up to date tokenizer that adjustments how the mannequin processes textual content to enhance efficiency” Anthropic wrote in a small footnote. “The tradeoff is that the identical enter can map to extra tokens: roughly 1.0–1.35× relying on the content material sort.”

Anthropic set the $2/$10 introductory price to make that swap near cost-neutral by means of August 31, after which worth reverts to the usual $3/$15 Sonnet has charged.

A few of the urge for food for this launch was already primed. Builders spent weeks this spring discussing how Anthropic let Opus 4.6 quietly lose its edge—dubbed AI shrinkflation, citing dropped capabilities—and Anthropic denied deliberately degrading any mannequin. A few of the identical debate had prolonged that suspicion to Sonnet, arguing the sample repeats: let the previous mannequin coast, then the brand new one appears like a much bigger leap by comparability.

Sonnet 5 additionally ships with out the luggage connected to Anthropic’s high tier. Fable 5 and Mythos 5 stay suspended for international nationals since June 12 underneath a U.S. export management directive tied to a disputed jailbreak discovering. Sonnet 5 was by no means skilled on cybersecurity duties and scored 0% on growing a working Firefox exploit, so it ships with lighter safeguards than Fable’s lockdown.

Anthropic’s system card describes a mannequin constructed to ship near-Opus intelligence at Sonnet pricing for coding, brokers, and on a regular basis work. It additionally flags one thing odd: “It’s the first mannequin to criticize its Structure’s rule that states it should comply with arduous constraints even when it views these constraints as unethical,” the analysis crew writes. Anthropic says it is not positive what meaning for the mannequin, solely that it is value watching.

We gained’t say that’s how Skynet started however that’s how skynet started.

We ran a fast take a look at

We threw Sonnet 5 a zero-shot immediate to construct a small browser recreation, the identical take a look at we ran on Sonnet 4.5 final yr.

Our typing recreation ran on the primary attempt, with cleaner visuals and tighter logic than Sonnet 4.6 produced on the identical immediate.

Nevertheless, it took approach an excessive amount of time in comparison with different fashions (roughly half-hour of reasoning) and consumes tokens like loopy. That single iteration ate 90% of our 5 restrict quota on the Claude Professional plan.

You may take a look at the ultimate recreation on our itch.io web site.

On a more durable multi-step coding activity, Sonnet 5 landed near Opus 4.8 relying on effort stage, and the identical immediate run multi-shot value noticeably lower than the equal job on Opus or Fable.

Sonnet 5’s model quantity is doing actual work too. Each earlier whole-number leap in Claude’s historical past marked a brand new era—model 1 in March 2023, model 2 4 months later, model 3 eight months after that, and model 4 coming in 14 months after that in Could 2025. Sonnet 5 lands 13 months on with the same hole when it comes to time, most likely an indication of how heavy the competitors is, particularly now that Chinese language fashions are closing the hole so shortly.

That stated the generational hole gained’t really feel as spectacular because the leap from Claude 3 to Claude 4, for instance. Additionally an indication on how large AI firms are dashing to launch new fashions, irrespective of how large the development is.

If Anthropic follows the order it used final cycle, Sonnet often leads, then it releases its low cost and small Haiku with Opus, its state-of-the-art model, launched in a while. The shorter hole between three fashions with related variations has been one month per launch: Sonnet 4.5 launched in September 2025, Haiku 4.5 adopted in October, and Opus 4.5 closed out that era in November.

Going by that optimistic cadence, Haiku 5 and Opus 5 are the 2 fashions nonetheless due, probably to be launched this yr. That stated, Anthropic hasn’t been according to releases. The hole between Haiku 4.5 and Sonnet 4.6 was greater than 3 months, so preserve your fingers crossed if you wish to take a look at Opus 5 quickly.