Claude Sonnet 5 Launches: Near-Opus Performance at a Fraction of the Cost

Anthropic has rolled out Claude Sonnet 5, a new version of its midsize AI model built to handle agentic work — planning multi-step tasks, operating tools like browsers and terminals, and completing jobs autonomously — at a fraction of the cost previously required for that level of performance. The release lands as every major AI lab races to prove its models can work independently rather than just answer questions in a chat window.

Agentic AI Becomes the Industry Standard

Sonnet 5 arrives just weeks after competing launches from OpenAI and Google that followed the same script. OpenAI introduced GPT-5.6 Sol in preview days earlier, its most agentic model so far, with the ability to divide work across multiple subagents to handle longer autonomous tasks. Google's Gemini 3.5 Flash, launched in May, made a similar pivot — moving away from a conversational assistant and toward a tool designed to plan, build, and iterate on real work with minimal human input.

That pattern signals where the competition is actually headed: agentic capability is no longer a differentiator on its own. The real battleground is cost and reliability — which company can deliver dependable autonomous performance at the lowest price.

What Sonnet 5 Improves On

Compared with its predecessor, Sonnet 4.6 (released in February), Sonnet 5 shows meaningful gains in reasoning, tool use, software coding, and general knowledge work. Anthropic also says the model is better at staying on task: testers found it completes complex, multi-step jobs that earlier versions would abandon partway through, and that it reviews its own output for errors without being explicitly told to.

One example shared by a Zapier engineer involved a two-part automation task — updating Salesforce account tiers, then sending a launch announcement to enterprise contacts — that Sonnet 5 carried through start to finish without stalling, a point Anthropic highlighted as evidence of its reliability for everyday business automation.

Pricing and Availability

As of Tuesday, Claude Sonnet 5 is the default model for both free and Pro plans and is available across every subscription tier.

Pricing is set at $2 per million input tokens and $10 per million output tokens through August 31. After that date, input pricing rises to $3 per million tokens while output pricing stays at $10 per million.

That positions Sonnet 5 as cheaper than Anthropic's own flagship Opus 4.8, as well as OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro. It remains more expensive than Google's Gemini 3.5 Flash, which currently holds the low-cost end of the market.

How Sonnet 5 Performs Against Opus 4.8

On agentic coding, Sonnet 5 closes much of the gap with Opus 4.8 compared to its predecessor, though Opus still leads. On a knowledge-work benchmark, however, Sonnet 5 edges slightly ahead of Opus 4.8 — notable given Opus's reputation for handling the hardest problems, including nuanced judgment calls and deep research tasks.

Anthropic frames the two models as complementary rather than competing: Opus 4.8 remains the better choice when accuracy is the top priority, while Sonnet 5 gives developers a markedly higher-quality option at a lower price point than was previously available. The company positions this as a dial users can adjust — trading cost for performance depending on the task.

Safety and Reliability

Anthropic reports that Sonnet 5 shows a lower rate of "undesirable behaviors" than Sonnet 4.6, including reduced cooperation with misuse attempts and less deceptive behavior. The model is also better at identifying and refusing malicious requests, including prompt-injection attempts designed to hijack its instructions. Hallucination rates and sycophantic tendencies are both down compared to the previous version.

Lovable co-founder Fabian Hedin pointed to this as a key strength for developers building tools used by large numbers of people, noting that a model's ability to decline unsafe requests is just as important as its ability to execute tasks well.

Despite these gains, Sonnet 5 doesn't match Opus 4.8 or Claude Mythos Preview on alignment-related safety evaluations. Anthropic specifically notes that Sonnet 5 has a much lower capability for dangerous cybersecurity tasks than its current Opus models — a limitation the company frames as intentional given the model's broader, lower-cost deployment.

Frequently Asked Questions

Is Claude Sonnet 5 free to use? Yes. As of its launch, Sonnet 5 is the default model on both Anthropic's free and Pro plans, and it's accessible across every subscription tier.

How does Claude Sonnet 5 compare in price to Opus 4.8? Sonnet 5 is cheaper than Opus 4.8, priced at $2 per million input tokens and $10 per million output tokens through August 31, after which input pricing increases to $3 per million tokens.

Is Claude Sonnet 5 better than Opus 4.8? Not across the board. Opus 4.8 still leads on agentic coding and is Anthropic's recommended model for tasks requiring the highest accuracy, but Sonnet 5 slightly outperforms it on a knowledge-work benchmark and offers significantly lower cost.