# Claude Sonnet 5.5 tightens Anthropic's grip on the mid-tier model market

> Anthropic held Sonnet pricing flat while cutting real task costs up to 30% with Sonnet 5.5.

*Anthropic is keeping Sonnet's rate card flat while cutting the token cost of running it, a pressure move on rivals selling cheaper mid-tier models.*

By Bez · InsidersFeed
Canonical: https://insidersfeed.com/news/claude-sonnet-5-5-anthropic-mid-tier-pricing

Anthropic used its second launch in six days to make a pricing argument rather than a pricing cut. Claude Sonnet 5.5, released on 28 September, keeps the exact rate card of its predecessor — $2 per million input tokens, $10 per million output tokens, 20 cents per million cache reads — while claiming a real-world cost drop of up to 30% per task, according to [Anthropic](https://www.anthropic.com/claude/sonnet). That is a different lever from a straightforward price cut, and one that is harder for rivals to match on a spec sheet.

The mechanism matters for buyers watching margins: Anthropic says the saving comes from the model needing fewer tokens and fewer tool calls to finish a job, not from underpricing capacity. For enterprise customers running high volumes through Claude Code or API integrations, that is a cost lever independent of the headline rate, and one that compounds at scale.

Anthropic pitched Sonnet 5.5 as the workhorse tier: strongest on well-scoped everyday tasks such as fixing bugs and producing documents, slides and spreadsheets, leaving Opus 5.5, launched a week earlier, as the reasoning-heavy flagship. Speed backs that positioning: Sonnet 5.5 runs more than 30% faster than Sonnet 5, Anthropic's fastest Sonnet yet, which matters for products built around interactive latency.

The benchmark gap to the flagship has narrowed sharply. Sonnet 5.5 scored 1844 on GDPval-AA v2.1, within two points of Opus 5.5's 1846 and far above Sonnet 5's 1449, plus 80.1% on OSWorld 2.1 and 64.5% with tools on Humanity's Last Exam, per Anthropic's own figures. A mid-tier model closing that far on a flagship narrows the case for always paying for the top tier.

Availability is broad from day one: Amazon Web Services, Google Cloud and Microsoft Azure all carry the model as claude-sonnet-5-5, alongside Claude Code, giving enterprise buyers already committed to one of those clouds an immediate path to adopt it without a procurement delay.

Anthropic is also using the launch to sell trust to regulated buyers. Sonnet 5.5 is the first Sonnet released with the cyber safeguards normally reserved for Anthropic's most capable models, and [SiliconANGLE](https://siliconangle.com/2026/09/28/anthropic-debuts-claude-sonnet-5-5-running-30-faster-than-the-previous-generation-ai-model/) reported its output carries an invisible watermark added for compliance with rules including the EU AI Act. In analysis terms, that combination is aimed at buyers who need compliance features without paying flagship prices.

Haiku 5.5 remains outstanding. Anthropic's Opus 5.5 announcement had promised both Sonnet 5.5 and Haiku 5.5 within weeks, and completing the cheapest tier will determine how far down-market this pricing squeeze reaches, and how much pressure it puts on rivals' budget-tier offerings, which is where volume buyers make their platform decisions.

For now, the comparison that matters to buyers is the one Anthropic itself invited: a mid-tier model priced like last year's mid-tier model, but benchmarking close to this year's flagship. That is the argument competitors selling on raw price per token will have to answer.

## Key takeaways

- Sonnet 5.5 keeps Sonnet 5's rate card but needs fewer tokens and tool calls, cutting real cost per task up to 30%.
- It scores 1844 on GDPval-AA v2.1, nearly matching flagship Opus 5.5's 1846.
- It ships with Opus-grade cyber safeguards, unusual for a mid-tier model this early.

## Sources

- [Claude Sonnet](https://www.anthropic.com/claude/sonnet) — Anthropic, 2026-09-28
- [Anthropic debuts Claude Sonnet 5.5 running 30% faster than the previous-generation AI model](https://siliconangle.com/2026/09/28/anthropic-debuts-claude-sonnet-5-5-running-30-faster-than-the-previous-generation-ai-model/) — SiliconANGLE, 2026-09-28