Is Claude Sonnet 5.5 Really 30% Cheaper Per Task?

Short answer

It can be true, but nothing gets cheaper on its own. The per-token price didn't change, so any saving comes from how many tokens Sonnet 5.5 uses, and that depends on the effort level you pick.Anthropic launched Claude Sonnet 5.5 on September 28, 2026, at the same API price as Sonnet 5: $2 per million input tokens and $10 per million output tokens. Anthropic says that in its own testing the model costs up to 30% less per task because it needs fewer tokens. Artificial Analysis ran both models at max effort on its Intelligence Index and got the opposite result: $7.60 per task for Sonnet 5.5 and $5.09 for Sonnet 5.

Source: Introducing Claude Sonnet 5.5 — Anthropic · 2026-09-29; Claude Sonnet 5.5 (max with fallback) — Artificial Analysis · 2026-09-29; Claude Sonnet 5 (max) — Artificial Analysis · 2026-09-29

What Anthropic actually promised

Read the wording closely. Anthropic's launch page says Sonnet 5.5 "typically needs far fewer tokens to do the same work" and that, in Anthropic's testing, it costs "up to 30% less per task than its predecessor." VentureBeat reports that Anthropic credits the saving to fewer tokens and fewer tool calls. It also reports Anthropic's claim that Sonnet 5.5 at Low or Medium effort can beat Sonnet 5's best score at roughly one-tenth the cost per task. The model has three effort settings: Low, Medium and High. Anthropic says Claude Code and its consumer apps default to Medium, while the Claude Platform (the API) defaults to High.

Source: Introducing Claude Sonnet 5.5 — Anthropic · 2026-09-29; Anthropic launches Claude Sonnet 5.5 with 30% cost reduction per-task — VentureBeat · 2026-09-29

A hand-drawn person working through the guide's check

What the independent test found

Artificial Analysis tested Sonnet 5.5 at max effort. The model scored 56 on its Intelligence Index, 18 points above Sonnet 5, and it used about 193k output tokens per task. Artificial Analysis says that is the highest token use it has measured. Because of those extra tokens, a task cost $7.60, about 50% more than Sonnet 5's $5.09. Low effort looks very different. Sonnet 5.5 on Low cost $0.41 per task and scored 36, just below Sonnet 5's max-effort score of 38. OrcaRouter, which highlighted these numbers, says effort levels don't carry over between models: Low on 5.5 doesn't think the same amount as Low on 5, so you have to re-test your settings.

Source: Claude Sonnet 5.5 reaches #2 on the Artificial Analysis Intelligence Index — Artificial Analysis · 2026-09-29; Claude Sonnet 5.5 (low with fallback) — Artificial Analysis · 2026-09-29; Claude Sonnet 5 (max) — Artificial Analysis · 2026-09-29; Claude Sonnet 5.5: The 30% Cost Claim vs $7.60 Per Task — OrcaRouter · 2026-09-29

Our take: both numbers can be true, and "up to 30%" is doing a lot of work in Anthropic's version. "Up to" means the best case, not the typical case. Anthropic's figure comes from its own tests, while the independent measurement at max effort went the other way. This isn't a scandal. It's a model that thinks a lot when you let it. What bothers us is the headline framing: many people read "30% cheaper" as a price cut, and it isn't one. If you just swap the model name in your API calls and keep the High default, don't expect your bill to shrink until you've measured it.

If you use the Claude app, cost means usage limits

Pro and Max subscribers don't pay per token. For them the relevant cost is how quickly they hit their usage limit. Claude's Help Center says effort level is one of the things that affects how much usage you consume, and the apps default Sonnet 5.5 to Medium. You can change the model, effort and thinking settings yourself.

Source: How do usage and length limits work? — Claude Help Center · 2026-09-29; Introducing Claude Sonnet 5.5 — Anthropic · 2026-09-29

Your usage is affected by several factors, including the length and complexity of your conversations, the features you use, which Claude model you're chatting with, and the effort level you've selected.

Claude Help Center, “How do usage and length limits work?” (read 2026-09-29)

Source: How do usage and length limits work? — Claude Help Center · 2026-09-29

Why this matters: Anthropic's own help page says effort level affects how fast you use up your allowance. So if you raise Sonnet 5.5 to High for everything, you should expect to hit your limit sooner than on the Medium default. Nothing is automatic here.

Source: How do usage and length limits work? — Claude Help Center · 2026-09-29

How to actually get the savings

API users: the Platform defaults to High, so set the effort level yourself instead of relying on the default.
Run your own tasks at Low, Medium and High, and compare the total cost per completed task, not the price per token.
Don't reuse your Sonnet 5 effort settings. Test them again on 5.5.
App users: try Medium first and switch to High only for work that really needs it.

Overrated: the "30% cheaper" headline. Underrated: Low effort. Artificial Analysis measured $0.41 per task on Low, and the score came within a couple of points of Sonnet 5 at max effort. That is where the real bargain is for routine work. Max effort is where it disappears.

Optional: see third-party seller listings for Claude subscriptions. This is a reseller listing, not official access and not an endorsement.. That link carries our partner code Sk52g, which is how a purchase is attributed to this site. Promotion code UPD3S.

Sources (all read 2026-09-29)

Claude · Compare price & terms →

Sonnet 5.5 Is '30% Cheaper'? Not at Max Effort

What Anthropic actually promised

What the independent test found

If you use the Claude app, cost means usage limits

Check the price before you renew