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

Anthropic has released Claude Sonnet 5, a mid-tier model that rivals its flagship Opus 4.8 on key benchmarks while costing up to 60% less per token. The launch comes as Anthropic races toward what could be the most scrutinized IPO in tech history, with a valuation now approaching $1 trillion.

Anthropic has released Claude Sonnet 5, positioning it as a high-performance, cost-efficient alternative to its flagship Opus 4.8 model. The launch is strategically timed as the company accelerates toward an IPO that will put its sky-high private valuations to the ultimate public test.

Pricing and Availability

Sonnet 5 is now the default model for Free and Pro plan users, and is also available on Max, Team, and Enterprise tiers.

Introductory API pricing is set at:

$2 per million input tokens and $10 per million output tokens through August 31
Rising to $3 input / $15 output after that date
Still well below Opus 4.8's $5 input / $25 output pricing

The strategic logic is clear: broaden developer adoption at accessible price points while building the kind of usage numbers that will read well in an S-1 filing.

Benchmark Results: Closing the Gap on Opus 4.8

Across every disclosed evaluation, Sonnet 5 makes substantial gains over its predecessor, Sonnet 4.6, and in several cases matches or exceeds Opus 4.8:

SWE-bench Pro (agentic coding): Sonnet 5 scores 63.2% vs. Sonnet 4.6's 58.1% and Opus 4.8's 69.2%
Terminal-Bench 2.1: Sonnet 5 at 80.4% vs. 67.0% (Sonnet 4.6) and 82.7% (Opus 4.8)
Humanity's Last Exam (reasoning with tools): Sonnet 5 scores 57.4% — essentially matching Opus 4.8's 57.9%
OSWorld-Verified (computer use): 81.2%, up from 78.5%
GDPval-AA v2 (knowledge work): Sonnet 5 scores 1,618, edging past Opus 4.8's 1,615

The pattern is consistent: Sonnet 5 doesn't inch forward — it vaults into performance territory that substantially overlaps with Anthropic's most expensive model, at roughly 60% lower cost per token.

Agentic AI: Finishing the Job

Anthropic describes Sonnet 5 as "the most agentic Sonnet model yet," reflecting the industry's shift toward AI systems that autonomously plan, use tools, and execute multi-step workflows — not just answer questions.

Early access partners reported a critical reliability improvement:

"With Claude Sonnet 5, agents stay on plan, follow our conventions, and ship clean multi-step changes, all at an efficient cost." — Sualeh Asif, co-founder, Cursor

"It used to stall halfway" — Daniel Shepard, Senior Engineer, Zapier, describing a two-part Salesforce and email automation task that now completes end to end.

This reliability gap — models that get 80% through a task before failing — has been a key blocker for production agentic deployments. Completing the full workflow changes the economics of enterprise automation entirely.

Anthropic also introduced cost-performance curves allowing developers to tune effort levels across Sonnet 5 and Opus 4.8, reflecting growing enterprise sophistication in AI consumption.

The Tokenizer Caveat

One detail buried in the footnotes warrants attention: Sonnet 5 uses an updated tokenizer, similar to the change introduced with Opus 4.7.

The same input can map to roughly 1.0 to 1.35× as many tokens depending on content type. Anthropic says introductory pricing is calibrated to be "roughly cost-neutral," but high-volume enterprise users should benchmark their specific workloads before assuming stable bills.

Safety Profile: Better Than Its Predecessor, Behind Its Flagship

Anthropic's safety disclosures present a nuanced picture:

Sonnet 5 shows lower hallucination and sycophancy rates than Sonnet 4.6
Improved refusal of malicious requests and greater resistance to prompt injection in agentic contexts
Scores lower (safer) than Sonnet 4.6 on Anthropic's automated behavioral audit

However, Sonnet 5 shows "somewhat higher rates of misaligned behavior" compared with Opus 4.8 and Anthropic's restricted Mythos-class models. On a Firefox 147 exploit development benchmark created with Mozilla, Sonnet 5 scored 0.0% working exploits — but showed a 13.2% partial success rate vs. 8.8% for Sonnet 4.6, while Opus 4.8 reached 68.8% and Mythos 5 hit 88.4%.

As a result, Sonnet 5 launches with cyber safeguards enabled by default, mirroring those on Opus 4.7 and 4.8. Organizations in Anthropic's Cyber Verification Program automatically receive equivalent access without reapplying.

The IPO Context: From $14B to $47B in Revenue

Sonnet 5 lands at a pivotal moment in Anthropic's trajectory. The company confidentially filed its IPO prospectus with the SEC in early June, in what CNBC has called "the most scrutinized public offering in tech history."

The financial acceleration has been staggering:

February 2025: $30B raised at a $380B valuation, with $14B annualized revenue — up tenfold in each of the prior three years
Late May 2025: $65B Series H closed at a $965B post-money valuation, co-led by Altimeter Capital and Sequoia Capital, with a revenue run rate crossing $47B

"The number that will either validate or collapse the entire narrative the private markets have been pricing for three years won't be the valuation or revenue — it'll be gross margin." — Harrison Rolfes, Analyst, PitchBook

Sonnet 5 serves a dual purpose: genuine capability gains for developers, and a compelling data point for Anthropic's IPO narrative — that it can deliver flagship-level performance at mass-market prices, at scale.