Claude Sonnet 5.5
Last updated: October 8, 2026
Sonnet 5.5 is the agentic work tier in Anthropic's current Claude lineup: the release positions it — with Opus 5.5 — to drive complex agentic coding, while Haiku 5.5 handles the high-frequency work beneath it.[1]
On October 7, 2026, Sonnet 5.5 got cheaper in one specific place — cache reads were halved — and the change shipped alongside the Haiku 5.5 release. Below: the price table, what moved, where Sonnet 5.5 stands on benchmarks, and how the two tiers divide work.
Prices
| Price per 1M tokens | Claude Sonnet 5.5 | Claude Haiku 5.5 |
|---|---|---|
| Cache reads | $0.10 — halved from $0.20 on Oct 7, 2026 | $0.01 / $0.05 (prompt ≤100k / prompt >100k) |
| Cache writes | $2.50 | $0.125 / $0.625 (prompt ≤100k / prompt >100k) |
| Input | $2.00 | $0.10 / $0.50 (prompt ≤100k / prompt >100k) |
| Output | $10.00 | $0.50 / $2.50 (prompt ≤100k / prompt >100k) |
Prices per million tokens, from Anthropic's October 7, 2026 release.[1] The stated Sonnet 5.5 change is the cache read cut, from $0.20 to $0.10. Haiku 5.5's rates are split by prompt length (≤100k and >100k tokens); Sonnet 5.5 is listed at flat rates.
What changed on October 7, 2026
- Cache reads halved. Sonnet 5.5 cache reads went from $0.20 to $0.10 per million tokens. Anthropic says the change makes agentic work around 20% cheaper to run — cache-heavy agent loops are where that lands.[1]
- Haiku 5.5 launched the same day. The release frames the two as a tier system rather than alternatives — Sonnet for the complex loop, Haiku 5.5 for the bounded calls inside it.
- Subscription plans gained API credits. Monthly credits were added: $100 on Max 5x, $200 on Max 20x, and up to $500 pooled for Team plans — usage credits, not per-token rate changes.
- SDKs added computer use and browser use. The Claude Python and TypeScript SDKs now ship computer use and browser use as beta capabilities.
How Sonnet 5.5 and Haiku 5.5 split the work
The release's division of labor is direct: complex agentic coding is Sonnet 5.5 / Opus 5.5 territory. Haiku 5.5 is named for summaries, compactions, database queries, classification requests, live customer support, browser use — and explicitly as a coding subagent alongside Sonnet 5.5 and Opus 5.5.[1]
The economics point the same way. Input is $2.00 against Haiku 5.5's $0.10 per million tokens at up to 100k prompts; output is $10.00 against $0.50. Routing bounded, high-frequency calls down to Haiku 5.5 keeps the expensive tier for work that needs it — and the same-day cache cut makes the expensive tier cheaper to hold open for exactly that role. The full head-to-head, including all nine shared data points, is on the Haiku 5.5 vs Sonnet 5.5 comparison.
Where Sonnet 5.5 stands on benchmarks
In the shared October 7 table, Sonnet 5.5 is the ceiling among the four models listed — Haiku 5.5, Haiku 4.5, GPT-6 Luna, and itself:
- Terminal-Bench 4.0 (agentic coding): 70.6% — against Haiku 5.5's 39.2% and GPT-6 Luna's 16.4%; the widest coding gap in the table.
- OSWorld 2.1 (computer use, offline subset): 83.9% — the highest score in the table, above Haiku 5.5's 72.4%.
- Humanity's Last Exam: 56.9% without tools, 64.5% with tools — both table-leading.
- FrontierCode 1.1 (Main): 52.1% — listed at the Xhigh effort setting.
- GDPval-AA v2.1: 1840 and AA-Briefcase v1.1: 1824 — the top scores on both knowledge-work rows.
The full four-model table, row by row, is on the Haiku 5.5 benchmark page.
Not yet confirmed from this source: Sonnet 5.5's context window, rate limits, throughput, and per-effort-setting scores. The release page does not publish them; they are labeled unpublished here rather than estimated.
Frequently asked questions
Did Claude Sonnet 5.5 get cheaper on October 7, 2026?
On caches, yes, per Anthropic: cache reads were halved from $0.20 to $0.10 per million tokens, and the release says that makes agentic work around 20% cheaper to run. The list prices are $2.00 input and $10.00 output per million tokens. If you run cache-heavy agent loops, the cache change is the one that moves your bill.
Is Sonnet 5.5 a brand-new model?
Nothing on the October 7 release page frames it as new — that release covers the changes that shipped alongside the Haiku 5.5 launch. What is verifiable from it: Sonnet 5.5 exists with these published prices and these benchmark scores. Its own launch details are Not yet confirmed on this page rather than assumed.
What are the new subscription API credits?
Anthropic added monthly API credits for subscription plans: $100 on Max 5x, $200 on Max 20x, and up to $500 pooled for Team plans. They are usage credits, not a change to the per-token rates.
Should I use Sonnet 5.5 or Haiku 5.5?
It splits by role, and Anthropic's release is explicit about it: Sonnet 5.5 (with Opus 5.5) for complex agentic coding, Haiku 5.5 for high-frequency bounded work — summaries, compaction, classification, database queries, live support — and as a coding subagent. The price juxtaposition explains the design ($2.00 vs $0.10 input per million tokens), and the benchmark gap explains why the top tier still exists.
Sources
- Anthropic — Claude Haiku 5.5 (official release page) Published October 7, 2026; where the same-day Sonnet 5.5 pricing changes and the shared benchmark table appear.
Sonnet 5.5 figures on this page come from Anthropic's official Claude Haiku 5.5 release page, published October 7, 2026 — where the same-day Sonnet pricing changes and the shared benchmark table appear. VerdictApi did not use third-party estimates; anything that page does not state is labeled Not yet confirmed.
Last updated: October 8, 2026