Claude Sonnet Explained: Features, Pricing, vs Opus

Last Updated: September 25, 2026

Summary: Claude Sonnet 5, released June 30, 2026, is Anthropic's mid-tier model built for agentic work — planning, using tools, and running autonomously at a fraction of flagship pricing. It costs $3 per million input tokens and $15 per million output tokens, scores close to Claude Opus on knowledge-work benchmarks, and remains the default model on Claude's free and Pro plans.

If you've opened Claude recently, there's a good chance Sonnet is the model actually answering you. Claude Sonnet is Anthropic's mid-tier model line, sitting between the fast, low-cost Haiku and the top-tier Opus, and as of Sonnet 5 it's been rebuilt specifically for agentic work: planning multi-step tasks, using tools like browsers and terminals, and running with less hand-holding than earlier versions needed. This guide covers what Claude Sonnet actually is, how the current version stacks up against Opus, what it costs, and how the naming has evolved through 3.5, 3.7, and now 5.

💡 Not sure which AI tool is actually right for your business?

Get the free guide, Which AI Tool Should You Actually Use? — a straightforward breakdown of the leading AI tools to help you pick the right one for your needs. Subscribe to AI Business Weekly for the guide, plus daily coverage of AI trends, acquisitions, and product launches.

Get the Free Guide →

Claude Sonnet is the middle tier in Anthropic's three-model lineup, positioned between Claude Haiku (fastest, cheapest) and Claude Opus (most capable, most expensive). Anthropic's own announcement describes the current version, Claude Sonnet 5, as "the most agentic Sonnet model yet" — built to make plans, operate tools, and complete autonomous multi-step work at capability levels that used to require a larger, pricier model.

Sonnet 5 became the default model across Claude's free and Pro plans starting June 30, 2026, according to TechCrunch's coverage of the launch, replacing Sonnet 4.6. Anthropic also reports lower rates of hallucination, deception, and sycophantic behavior compared to that predecessor, alongside the capability jump.

Claude Sonnet has gone through several distinct versions, and search interest still splits across them: Claude 3.5 Sonnet and Claude 3.7 Sonnet each retain meaningful search volume even after being superseded, since plenty of documentation, integrations, and older comparisons still reference them by name.

Anthropic skipped straight from naming conventions like "Claude 3.7 Sonnet" to "Claude Sonnet 5," dropping the version-number-first format, which is part of why searches for the older names haven't fully died down — people are still searching by the naming pattern they learned first. It's a naming quirk worth knowing if you're troubleshooting an older integration: a tool or tutorial referencing "Claude 3.5 Sonnet" by name is describing a model from an earlier generation, not a typo or a different product line.

The honest answer is closer than the price difference suggests. Per Anthropic's own comparison, Sonnet 5's "performance is close to that of Opus 4.8, but at lower prices," and on some knowledge-work benchmarks, Sonnet 5 actually edges Opus out. TechCrunch's testing found Sonnet 5 scoring 63.2% on agentic coding benchmarks versus Opus's 69.2% at the time, a real but modest gap given the price difference.

That gap has moved again since. Anthropic released Claude Opus 5.5 on September 22, 2026, as "the first model in a new 5.5 model family," according to MacRumors' coverage of the launch. Opus 5.5 surpassed Anthropic's own Fable 5.1 model on agentic coding, knowledge work, computer use, and multidisciplinary reasoning, while costing 40% less to run than the previous Opus generation and generating output over 30% faster. As of this writing, Anthropic has said Claude Sonnet 5.5 and Claude Haiku 5.5 are still in development and expected "in the coming weeks" — meaning Sonnet 5 remains the current Sonnet-tier model even after Opus moved up a full generation.

"Agentic" gets used loosely across the AI industry, so it's worth being specific about what it means for Sonnet 5. Rather than answering one prompt at a time and waiting for the next instruction, Sonnet 5 is built to take a goal, break it into steps, use tools like a browser or a terminal to execute those steps, and check its own work before handing back a result. Early users cited in TechCrunch's coverage of the launch noted the model "completes complex multi-step tasks end-to-end" and "checks its own output without explicitly being asked" — a meaningful behavioral shift from a model that simply responds to whatever the last message asked.

In practice, this shows up most clearly in coding and research workflows: instead of generating one function and stopping, Sonnet 5 can plan out a multi-file change, implement it, run tests, and correct its own mistakes within a single extended task. That's the specific capability Anthropic is pointing to when it says Sonnet 5 approaches Opus-level performance at a fraction of the cost — the gap has narrowed most on exactly this kind of multi-step, tool-using work, less so on tasks requiring deep, sustained judgment across a very long context.

Sonnet's practical sweet spot is high-volume, moderate-complexity work where cost compounds fast. Developers use it for day-to-day coding: writing functions, reviewing pull requests, and running multi-step debugging sessions where an agentic loop genuinely saves time over manual back-and-forth. Content and research teams use it for drafting, summarizing, and multi-source synthesis where the task is substantial but not at the level of a full investigative report. Businesses running Claude through the API for customer-facing tools — support triage, internal documentation search, structured data extraction — tend to default to Sonnet specifically because per-token cost matters at volume in a way it doesn't for a single one-off task.

The common thread across all of these is repetition and scale. A task run once a month can absorb Opus pricing without much thought; a task run thousands of times a day cannot, which is exactly the gap Sonnet is built to close.

Claude Sonnet 5 launched with introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026, per TechCrunch's reporting on the launch. Standard pricing since then runs $3 per million input tokens and $15 per million output tokens. Anthropic notes Sonnet 5 uses an updated tokenizer, so the same input text can consume 1.0 to 1.35 times more tokens than it would have on a previous model — worth factoring in if you're estimating costs from older usage patterns.

For most individual users, Sonnet 5 access is included in Claude's Free and Pro subscription plans rather than billed per token, and per-token API pricing mainly matters for developers building on Claude directly. Our AI pricing guide breaks down subscription pricing across Claude and its competitors if you're comparing plans rather than API costs specifically.

Sonnet 5 is the better fit for most day-to-day work: coding tasks, drafting and editing, research summaries, and any agentic workflow where you're running many tasks and cost adds up fast. Opus 5.5 earns its higher price on the harder end of the spectrum — long, sprawling jobs like a codebase-wide migration, a deep audit, or research requiring sustained, complex judgment across many steps.

If you're not sure which you need, Anthropic's own cost-performance data suggests starting with Sonnet by default and reaching for Opus specifically when a task is failing on Sonnet or is complex enough that you'd rather not risk a lower-effort pass. For a broader look at how Claude compares to other AI assistants entirely, our ChatGPT vs Claude and Claude vs Perplexity comparisons cover that ground.

Claude Sonnet is Anthropic's agentic, mid-tier model, and as of Sonnet 5 it closed most of the capability gap with Opus while staying meaningfully cheaper. Sonnet 5 launched June 30, 2026 at $2/$10 per million tokens introductory pricing, now $3/$15 standard, and remains the default model on Claude's free and Pro plans even after Opus 5.5's September 2026 release moved the flagship tier forward again. For most everyday work, Sonnet is the practical choice; Opus is worth the premium specifically for the longest, hardest jobs.

What is Claude Sonnet used for?

Claude Sonnet is built for agentic, day-to-day work: coding tasks, planning and completing multi-step projects, using tools like browsers and terminals, and general writing or research work. It's designed to run more autonomously than earlier Claude versions, checking its own output rather than needing every step confirmed. This doesn't cover the hardest, most sprawling jobs, where Opus still has an edge. Most everyday Claude usage on the free and Pro plans runs on Sonnet by default.

Is Claude Sonnet better than Claude Opus?

Not better overall, but close on many tasks and meaningfully cheaper. Anthropic's own data shows Sonnet 5 performing near Opus's level on knowledge-work benchmarks and even ahead on some, while costing about a quarter as much per million tokens. This doesn't apply to the longest, most complex jobs, like a full codebase migration, where Opus 5.5 is built to hold up better over many steps. If cost matters and the task is moderate in complexity, Sonnet is usually the better starting point.

How much does Claude Sonnet cost?

Claude Sonnet 5 costs $3 per million input tokens and $15 per million output tokens as of standard pricing (introductory pricing through August 31, 2026 was $2 and $10). This is API pricing for developers; most individual users access Sonnet through Claude's Free or Pro subscription instead of paying per token. This doesn't include Anthropic's updated tokenizer, which can make the same input consume up to 1.35 times more tokens than on older models. See our AI pricing guide for subscription-level pricing comparisons.

What happened to Claude 3.5 and 3.7 Sonnet?

Both were superseded by later versions and are no longer Anthropic's current models, though they remain referenced in older documentation, integrations, and comparisons, which is why they still show up in searches. Claude Sonnet 4.6 followed those, then was itself replaced by Claude Sonnet 5 in June 2026. This doesn't mean older versions stopped working everywhere overnight, since some third-party tools and integrations take time to migrate. If you're building something new, Sonnet 5 is the current version to target.

Is Claude Sonnet 5 the newest Claude model?

No — Claude Opus 5.5 released in September 2026 as the newer flagship model, though Sonnet 5 remains the current version specifically in the Sonnet tier. Anthropic has said a Sonnet 5.5 update is in development. This doesn't mean Sonnet 5 is outdated for its purpose, since Anthropic explicitly built it as the practical, cost-efficient option rather than the top-capability one. Check Anthropic's own release page for the latest status before assuming either name is current.

By Sameer Khan

This article was AI-assisted, then reviewed by Sameer Khan before publishing.

Sameer Khan is the founder of AI Business Weekly. He has a background in research and advisory, working with HR leaders and executives across Canadian public-sector and enterprise organizations on research and AI adoption. He holds an MBA from the Ted Rogers School of Management and has spent nearly a decade in B2B sales across SaaS, research and advisory, and AI.