Anthropic releases Claude Sonnet 5.5, a faster and cheaper model that nearly matches Opus

Sonnet 5.5 is a faster, lower-cost model that Anthropic says is better at coding, completing scoped tasks, and creating polished documents, slides, and spreadsheets than its predecessor.

• Speed: More than 30% faster than Sonnet 5.
• Cost: Up to 30% less per task than Sonnet 5, and requires fewer tokens overall.
• Pricing: $2 per million input tokens, $10 per million output tokens, half of Opus 5.5's rate.

The catch: Sonnet 5.5 doesn't advance Anthropic's frontier capabilities. It's explicitly positioned as the practical, everyday option rather than the smartest model the company can build.

Anthropic research product manager Theo Chu explained the positioning directly to CNBC. "Sonnet is really for the cost-conscious customer where they might not need as much intelligence," Chu said.

In real life If you're using Claude for routine work like drafting emails, cleaning up spreadsheets, or fixing straightforward bugs, Sonnet 5.5 is built for exactly that, saving Opus-level power for harder problems.

"It might be routine tasks that just need execution, but don't need that judgment that Opus can bring," Chu added, describing the split Anthropic wants customers to make between its two main model tiers.

Despite costing half as much as Opus 5.5, Sonnet 5.5's benchmark scores land close enough to Opus that Android Authority described the gap as little more than rounding-error territory on many tasks.

Among the more unusual benchmarks Anthropic and reviewers pointed to: Sonnet 5.5 is capable of completing the classic video game Pokémon Red, a long-running informal test some AI labs use to demonstrate sustained multi-step reasoning and planning ability.

It's a quirky measure, but one that's become a real shorthand in the industry for whether a model can hold a long-term goal across many small decisions without losing the plot, a skill that matters just as much for spreadsheet work as it does for catching Pokémon.

Sonnet 5.5 is the first Sonnet model to ship with cybersecurity fallbacks and safeguards Anthropic previously reserved for its most capable models. The company says those cybersecurity capabilities represent a "large improvement" over the previous Sonnet generation.

It's also the first Claude model empowered to resist distillation attacks, a technique where outside parties try to extract or copy a model's underlying training by querying it extensively. Protecting against that matters more as Sonnet-tier models get cheap enough that copying one could otherwise be commercially attractive.

This release lands in the middle of a tense stretch for the industry. Sonnet 5.5 is Anthropic's second model launch since CEO Dario Amodei publicly urged AI companies to slow how quickly they improve their most advanced models, a call that stunned much of the tech sector and government officials when he made it earlier this month.

Because Sonnet 5.5 doesn't push Anthropic's capability frontier forward, the company says most of its alignment testing focused on "a targeted set of risks that apply to models of any capability level," rather than the more extensive review reserved for frontier releases like Opus 5.5.

"Focusing on alignment and safety has been a key part of our mission from the very beginning," Chu told CNBC. "This is something that we've always prioritized across our models."

That framing matters because the broader AI industry has faced intense scrutiny in recent weeks, with researchers repeatedly warning about the potential for advanced systems to cause serious, hard-to-predict harm as they grow more capable.

Sonnet 5.5's $2-per-million-input, $10-per-million-output pricing doesn't exist in a vacuum. OpenAI's GPT-6 Sol currently prices at the identical $2 input rate, with output priced slightly cheaper at $10 per million tokens as well, putting the two head to head on paper.

By the numbers: Google's Gemini 3.1 Pro runs $2 per million input tokens but $12 per million output tokens, making it modestly more expensive than Sonnet 5.5 for output-heavy work. Google's faster, cheaper Gemini 3.8 Flash undercuts all three at a promotional $0.75 input and $3.75 output rate, though that pricing is set to roughly double starting January 2027.

That puts Sonnet 5.5 squarely in the middle of a genuinely competitive mid-tier field, rather than standing out as a clear price leader. Anthropic's pitch isn't that Sonnet 5.5 is the cheapest option available, it's that the combination of price, speed, and near-Opus performance is hard to match line for line against any single competitor.

Sonnet's pricing isn't just a consumer-facing detail. Anthropic's business runs overwhelmingly on pay-per-token API usage rather than subscriptions, with roughly 72% of revenue coming from companies making direct API calls and only about 13% from Claude.ai subscriptions, according to industry analysis of the company's business model.

The big picture: that makes a fast, cheap, high-volume model like Sonnet 5.5 core to Anthropic's actual revenue engine, not a side offering. The company's annualized revenue run rate reportedly reached roughly $65 billion by July 2026, up from just $1 billion in December 2024, one of the fastest revenue growth trajectories ever recorded in enterprise software.

Anthropic has also reportedly filed confidentially for an IPO at a $965 billion valuation, and as of earlier this year its enterprise AI market share had reportedly surpassed OpenAI's for the first time. Keeping a cheap, fast, reliable model in the lineup is central to defending that lead against competitors chasing the same high-volume enterprise API business.

Sonnet 5.5 arrives less than a week after Anthropic introduced Opus 5.5, its most capable model, which set new performance and communication benchmarks when it launched. Anthropic also confirmed Monday that a cheaper Haiku 5.5 model is coming soon, which would round out a refreshed lineup across all three of its pricing and capability tiers.

Sonnet 5.5 is available now across every major platform offering Anthropic's models, including Amazon Web Services, Google Cloud, and Microsoft Azure, so existing Claude customers can switch to it immediately without waiting for a broader rollout.

Anthropic counts customers like Netflix, Spotify, KPMG, and Salesforce among its enterprise base, and the number of customers spending over $1 million a year on Claude reportedly doubled in just a few months earlier in 2026. A model like Sonnet 5.5, cheap enough for high-volume routine work but capable enough to trust with real tasks, is exactly what keeps that kind of enterprise spending scaling up rather than flattening out.