Anthropic released Claude Sonnet 5 on Tuesday, and most of the coverage is leading with the wrong headline. Yes, it’s a faster, smarter model. That’s expected with every release. The part that actually matters for anyone running a business is the price.
Sonnet 5 performs close to Anthropic’s flagship Opus 4.8 model on agentic tasks like coding, multi-step planning, and tool use, but it costs roughly 60 percent less to run. That gap is the story. It changes the math on what’s actually affordable to automate.
What Changed
Anthropic priced Sonnet 5 at $2 per million input tokens and $10 per million output tokens through August 31, 2026, stepping up to $3 and $15 after that. Opus 4.8, the company’s top model, runs $5 input and $25 output. So you’re looking at a model performing near the top of the market for somewhere around a third of the cost.
The benchmark numbers back up the positioning. On one agentic coding test, Sonnet 5 scored 63.2 percent, up from 58.1 percent on the prior Sonnet 4.6 release, and within striking distance of Opus 4.8’s 69.2 percent. On a knowledge work benchmark, Sonnet 5 actually came out slightly ahead of Opus 4.8. On Terminal-Bench, a measure of how well a model handles real command-line and tool-use tasks, Sonnet 5 jumped to 80.4 percent from 67.0 percent on the previous version.
None of that means Sonnet 5 replaces Opus 4.8 outright. Anthropic is straightforward about that: Opus is still the better choice when accuracy matters most and cost is secondary. But for the bulk of day-to-day automation work, businesses now have a model that gets most of the way to flagship performance without flagship pricing.
Built to Finish the Job, Not Just Start It
The bigger shift, at least from Anthropic’s own framing, is around follow-through. Early testers reported that Sonnet 5 completes complex, multi-step tasks where earlier Sonnet models would stall partway and hand control back to a human.
One example Anthropic shared came from a team that handed the model a two-part job: update account tiers in a CRM, then send a launch announcement to a list of contacts. The model carried it through start to finish without needing someone to pick up the second half manually. That’s a small example, but it points at a bigger pattern. The work AI tools can be trusted to run without supervision keeps expanding, and the cost of running that work keeps dropping.
For agencies and operations teams, that combination is what actually moves the needle. A model that’s smart but expensive gets used selectively. A model that’s nearly as smart and a third of the price gets built into everyday workflows.
Safety Did Not Take a Back Seat
Anthropic also reported that Sonnet 5 shows lower rates of problematic behavior compared to its predecessor, including less cooperation with misuse attempts, less deception, and fewer hallucinations. It ships with cyber safeguards turned on by default, similar to what’s used on Opus-class models, though less restrictive than the safeguards applied to Anthropic’s more locked-down Mythos-class models.
That last detail matters if you’ve been following the recent back-and-forth around Claude Fable 5, which was pulled from public access by a government directive shortly after launch. Sonnet 5 is not built on that same restricted architecture, and Anthropic specifically designed it with lower-risk cybersecurity capability so it could ship without the same complications.
Why This Matters Beyond the Tech Press
This launch lands while Anthropic is reportedly preparing for an IPO, in a market that’s watching closely to see whether private AI valuations hold up once they’re tested in public. A company gearing up for that kind of scrutiny doesn’t cut prices on a near-flagship product by accident. It’s a signal that the AI labs are now competing on cost as hard as they’re competing on raw capability, which is usually what happens once a technology moves from “impressive demo” to “operational tool businesses actually budget for.”
For business owners weighing whether AI automation is worth building into operations, that’s the practical takeaway. The conversation is no longer just “can the model do this.” It’s “can the model do this at a price that makes sense for the volume of work we actually have.” Sonnet 5 is Anthropic’s answer to that second question, and it’s a stronger answer than what existed four months ago when Sonnet 4.6 launched.
Curious how a model release like this changes what’s actually worth automating in your business? Book a strategy session and we’ll walk through it.
