The Complete Claude Sonnet 5.5 Guide
As of 2026-10-19, Claude Sonnet 5.5 (model ID claude-sonnet-5-5, released 2026-09-28) lists at input $2 / output $10 per million tokens; since 2026-10-07 cache reads cost $0.10 instead of $0.20 (0.05x the base input price), with cache writes and all other prices unchanged. It has a 1M context window and 128K maximum output, and input and output still match Sonnet 5. Anthropic says it runs 30%+ faster than its predecessor and costs up to 30% less for most work. Anthropic positions it as the tier that balances speed and intelligence; long-running agentic coding and knowledge work still belong to Opus 5.5. This page lays out the specs, pricing, default effort and the breaking changes in one go.
Updated 2026-10-22
Claude, GPT and Chinese models on one key, billed per token; check /models for callable models.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Specs and defaults
Input / output (per million tokens)
The launch post states that input and output are priced the same as on Sonnet 5; both generations carry identical numbers in the official price list: Sonnet 5's $2 / $10 is confirmed as the standard rate, and the increase to $3 / $15 once scheduled for 2026-09-01 will not happen.
Cache write 5-minute / 1-hour
Cache-write prices in the official list: $2.50 for 5 minutes, $4 for 1 hour; the official release note of 2026-10-07 says cache writes are unchanged. Cache writes are billed by cache duration, and the 5-minute and 1-hour rates differ. Cache reads have cost $0.10 since 2026-10-07 (previously $0.20), per the release notes and the prompt-caching section of the pricing page.
Context window / max output
From the official model table: a 1M-token context and 128K-token maximum output, with a reliable knowledge cutoff of June 2026.
Default effort
The official model table gives Sonnet 5.5 a default effort of high; in Claude Code and the official apps the default is medium, and effort can be set explicitly.
Where it sits in the family
The official launch post calls Sonnet 5.5 the second model released in the Claude 5.5 family (the first was Opus 5.5, released 2026-09-22; Haiku 5.5 followed on 2026-10-07). The official wording on the division of labor is that complex work requiring careful judgment goes to Opus 5.5, while everyday tasks with clear scope — fixing bugs, and producing polished documents, slides and spreadsheets — are where it is strongest. Price is parity, not a selling point: at launch the post said input $2, output $10 and cache reads at $0.20 cost the same as on Sonnet 5, and Batch is $1 / $5; since 2026-10-07 its cache reads cost $0.10, while input and output still match Sonnet 5. Sonnet 5 is not being deprecated either: as of 2026-10-08 the official deprecation table marks both rows Active. The official model table describes Sonnet 5.5 as the combination of speed and intelligence and leaves long-running agentic coding and knowledge work to Opus 5.5; its guidance is plain: if you are unsure which model to use, start with Claude Opus 5.5 for most workloads.
The vendor's own framing (released 2026-09-28)
The official model table describes Sonnet 5.5 as the combination of speed and intelligence, Opus 5.5 as the model for long-running agentic coding and knowledge work, and Fable 5.1 as the one for demanding reasoning and long-horizon agentic work; the guidance it gives is plain: if you are unsure which model to use, start with Claude Opus 5.5 for most workloads. The launch announcement says Sonnet 5.5 runs 30%+ faster and costs up to 30% less for most work. The Claude Code changelog is currently at 2.1.303, and since 2.1.284 Sonnet 5.5 (claude-sonnet-5-5) has been the default Sonnet choice in Claude Code on the Anthropic API — that is the default this client picks, not an API-wide default model. On 2026-10-07 Anthropic launched Claude Haiku 5.5 (claude-haiku-5-5), and the same announcement halves the price of Sonnet 5.5's cache reads; the company says this makes Sonnet 5.5 around 20% cheaper on most agentic work.
The Sonnet 5.5 timeline
Claude Opus 5.5 is released; the official Opus 5.5 page names Sonnet 5.5 and Haiku 5.5 as the rest of the family; both later released, on 2026-09-28 and 2026-10-07 respectively.
Claude Sonnet 5.5 ships at the same price as Sonnet 5; from Claude Code 2.1.284 the default Sonnet becomes it.
Official release notes: prompt cache reads on Sonnet 5.5 drop from $0.20 to $0.10 per million tokens, 0.05x the base input price instead of 0.1x, with cache writes and all other prices unchanged; Claude Haiku 5.5 launches the same day.
Confirmed vs not verified
Confirmed (verbatim in the official pages)
All of the following are officially confirmed and can be checked word for word in the official docs and price list: released 2026-09-28, model ID claude-sonnet-5-5, 1M / 128K, input $2 / output $10, cache reads $0.10 (since 2026-10-07, previously $0.20, per the release notes and the prompt-caching section of the pricing page), cache writes $2.50 and $4, Batch $1 / $5, priced the same as Sonnet 5 at launch, default effort (the official model table gives Sonnet 5.5 a default of high, while Claude Code and the apps default to medium), the list of breaking changes that shipped with the model, and the default-tier change in Claude Code 2.1.284. The speed, efficiency and score figures also come from the official pages, but those are the vendor's own claims and Anthropic's own testing, not a third-party audit. One further item is not a vendor statement but a measurement taken here on 2026-09-29: claude-sonnet-5-5 is available to call on QCode.
Unverified and not published by the vendor
These points have not been verified, so don't plan around them: first, the figures of 30%+ faster and up to 30% less per task, and the claim that halving cache reads makes most agentic work around 20% cheaper, are the vendor's own measurements, and no third party has replicated them outside the official pages; second, comparison figures and screenshots reportedly circulating online have no first-hand source. One discrepancy on the pricing pages is worth remembering: when captured on 2026-10-08, both the main model-pricing table on the official pricing page and the price table on the Sonnet 5.5 overview page still printed $0.20 for Sonnet 5.5 cache reads, which disagrees with the $0.10 in the 2026-10-07 release notes and in the prompt-caching section of that same pricing page; on the 2026-10-19 re-check both places show $0.10, and this page follows the release notes. Haiku 5.5, previously listed here, launched on 2026-10-07 (claude-haiku-5-5) and is no longer unpublished.
How it stacks up against Sonnet 5 and Opus 5.5
vs Sonnet 5
At launch the official post said input $2, output $10 and cache reads at $0.20 were priced the same as on Sonnet 5; since 2026-10-07 Sonnet 5.5's cache reads cost $0.10 while Sonnet 5 stays at $0.20, and input and output are the same on both. The difference is efficiency and behavior: in Anthropic's own testing on Terminal-Bench 4.0, Sonnet 5.5 scores 70.6% and Sonnet 5 scores 10.3%, and the company says it runs 30%+ faster; no third party has audited these numbers. What actually demands code changes is the set of breaking changes that shipped with the model.
vs Opus 5.5
The two are a division of labour, not a ranking: Anthropic reserves Opus 5.5 for complex work that calls for careful judgment, including long-running agentic coding and knowledge work, while Sonnet 5.5 takes everyday tasks with clear scope. Converted from the official price list, Sonnet 5.5 is exactly half of Opus 5.5 on input and output ($2 / $10 vs $4 / $20). Since 2026-10-07 cache reads are $0.10 on Sonnet 5.5 and $0.20 on Opus 5.5; the pricing page applies the same cache-hit rule to Opus 5.5 and to Sonnet 5.5. Fast mode currently covers only Opus 5.5, Opus 5 and Opus 4.8, the latter two being a research preview, so Sonnet 5.5 is not part of it.
Five things to check when you switch over
The official migration guide states that five breaking changes affect code already running on Claude Sonnet 5: first, up-front thinking can be turned off with between_tools; second, forced tool use returns an error; third, thinking blocks are tied to the model and conversation that produced them; fourth, the Claude API and Google Cloud no longer accept the earlier computer_20251124 computer-use tool; fifth, the advisor tool rejects Claude Opus 4.8, Claude Opus 4.7 and Claude Sonnet 5 as advisors. One further change breaks no request but alters the shape of the response: text between tool calls comes back inside thinking blocks.
On QCode
claude-sonnet-5-5 has been available on QCode since 2026-09-29 (official list $2 / $10, cache reads $0.10 since 2026-10-07). To call it, copy a key in the console, use QCode's Anthropic-compatible endpoint, put claude-sonnet-5-5 in the model field and leave the rest as it is. The same key also runs claude-sonnet-5 and claude-opus-5-5: switching changes the model field only, the endpoint and the key stay the same. Billing is per token; each model's price is listed on /models. Want to try the new model for a few days first? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days.
Frequently asked questions
How is the claude-sonnet-5-5 ID written on each platform?
The ID is written differently per platform: claude-sonnet-5-5 on the Claude API, Google Cloud, Microsoft Foundry and Claude Platform on AWS, and anthropic.claude-sonnet-5-5 on Amazon Bedrock. The Claude API alias in the official model table is also claude-sonnet-5-5, with no date suffix, and the deprecation table marks it Active, with retirement no sooner than 2027-09-28.
How much does Sonnet 5.5 cost? Same as Sonnet 5?
Input and output are the same; cache reads are cheaper. The Sonnet 5.5 row in the official price list reads: input $2, a 5-minute cache write $2.50, a 1-hour cache write $4, output $10 per MTok; Batch is $1 / $5. Cache reads: the official release notes for 2026-10-07 say they dropped from $0.20 to $0.10 (0.05x the base input price), with cache writes and all other prices unchanged, while Sonnet 5's cache reads stay at $0.20. When captured on 2026-10-08, the cache-read cell in that price-list row still printed $0.20; on the 2026-10-19 re-check the model price table and the prompt-caching section of the pricing page both show $0.10, so the two agree. At launch the post said input, output and cache reads were all priced the same as Sonnet 5. One more: Sonnet 5's scheduled 2026-09-01 increase to $3 / $15 will not occur, per the price-list footnote.
How much faster and cheaper is it than Sonnet 5?
Anthropic says it runs 30%+ faster than the previous generation and costs up to 30% less per task than its predecessor. Those are the vendor's own claims and self-testing, with no third-party audit; the official pages carry other self-tested scores too, among them the Terminal-Bench 4.0 figures, 70.6% vs 10.3%, which are equally Anthropic's own testing. Input and output prices match Sonnet 5, and cache reads are lower since 2026-10-07 ($0.10 vs $0.20); the company says halving cache reads makes Sonnet 5.5 around 20% cheaper on most agentic work, again a vendor claim. Your actual saving depends on your own token usage and cache hits.
Do I need to change code when moving from Sonnet 5?
Possibly. The breaking changes the vendor lists include the thinking-related ones, and they are the easiest to hit: switching thinking off explicitly returns an error, and between_tools is accepted only at the low, medium and high effort levels — at xhigh or max it returns a 400 error, so at those levels you have to move to adaptive thinking.
What error does turning off thinking return?
The official migration docs give the error as follows: thinking.type.disabled is not supported for this model. For the lowest thinking setting use thinking.type.between_tools; to control thinking behaviour use thinking.type.adaptive together with output_config.effort, and stop sending thinking.type.disabled.
Can I call it on QCode right now?
Yes. claude-sonnet-5-5 has been on QCode's list of callable models since 2026-09-29. Your existing key still switches between claude-sonnet-5 and claude-opus-5-5 by changing the model field only. Billing is per token; for the current list and prices, go by /models.
Sources
Specs and change list: the Claude Sonnet 5.5 overview page and the model table in Anthropic's official docs (first crawled 2026-09-29; release date, model ID, context and maximum output, per-platform IDs, default effort and retirement date re-checked 2026-10-19). Pricing: the official pricing page, which supplies the per-row input, output and cache-write prices for Sonnet 5.5, Sonnet 5 and Opus 5.5, the Batch rates and the footnote cancelling the Sonnet 5 increase; the $0.10 cache-read price from 2026-10-07 comes from the 2026-10-07 entry in the Claude Platform release notes and the prompt-caching section of the pricing page (crawled 2026-10-08, re-checked 2026-10-19). Positioning and self-reported figures: the official launch post claude-sonnet-5-5 (the speed and cost claims, the Terminal-Bench 4.0 score from Anthropic's own testing) and the 2026-10-07 Claude Haiku 5.5 announcement (cache reads halved, the around-20% vendor claim). The error text and between_tools' effort limits: the official migration guide. The default-Sonnet change: the Claude Code changelog 2.1.284; the current latest version 2.1.303 comes from that same changelog. All of these pages were fetched and checked on 2026-10-19.
The newest same-price tier, ready today
Call claude-sonnet-5, claude-sonnet-5-5 (available on QCode since 2026-09-29) and claude-opus-5-5 from one endpoint, billed per token, ready the moment you sign up.
Claude Sonnet 5.5 breaking changes and errors
Anthropic released Claude Sonnet 5.5 on 2026-09-28, and the official overview states «Five breaking changes affect code already running on Claude Sonnet 5»; the vendor's migration guide separately lists «five settings that return a 400 error» — the two sets of five come from two different official documents, not one table. This section puts both kinds of official sentence next to the verbatim error strings: turning off up-front thinking, forced tool use, thinking-block binding, computer_20251124 no longer accepted, and advisor pairing, plus one change that raises no error but reshapes the response.
The five confirmed changes (with Sonnet 5.5 since 2026-09-28): (1) up-front thinking is turned off with between_tools — thinking itself cannot be disabled; (2) forced tool use is rejected; (3) thinking blocks are bound to the model and the session; (4) on the Claude API and Google Cloud the earlier computer_20251124 computer-use tool is not accepted; (5) the advisor tool rejects Claude Opus 4.8, Claude Opus 4.7 and Claude Sonnet 5 as advisors. The vendor separately notes one more change that fails no request but moves text between tool calls back into thinking blocks.
The error reads: "thinking.type.disabled" is not supported for this model. Use "thinking.type.between_tools" for the lowest thinking setting, or "thinking.type.adaptive" and "output_config.effort" to control thinking behavior. The lowest thinking setting the vendor documents is between_tools, accepted at low, medium and high effort. At xhigh or max it returns a 400 error, and there adaptive thinking is the answer: omit the thinking field or send adaptive. between_tools accepts no other field — display, budget_tokens or block_binding sent with it returns 400. And once between_tools is in use, effort cannot change mid-conversation.
Related reading
Claude Sonnet 5 complete guide
The same-priced previous generation Sonnet 5: specs, price list and selection advice.
Claude Sonnet 5 pricing and cost per token
Sonnet 5's per-unit prices, the cancelled 2026-09-01 increase and how caching is billed.
Claude Opus 5.5 complete guide
The tier the vendor reserves for complex work: $4 / $20, cache reads $0.20. Long-running agentic coding and knowledge work go to it as well.
Crypto payment guide
Nine coins, 56 networks, zero platform fee for AI coding API credits.
Prices and specs on this page were checked on 2026-10-19, and quotes and Claude Code details on 2026-09-29, against Anthropic's official docs and price list; changes on Anthropic's side may happen without notice. All speed, cost and score numbers are the vendor's own claims or self-testing, not a third-party audit, and are not a promise about your specific workload. Model availability is whatever /models and your own testing show.