News / Model Watch · Family rollout
Claude 5.5 Family Rollout Complete: the Full Rate Card in One Place
Three Tuesdays in a row finished the set: Opus 5.5 on Sep 22 at $4/$20, Sonnet 5.5 on Sep 28 at $2/$10 — confirmed as the regular price, with the Sep 1 plan to raise it to $3/$15 cancelled — and Haiku 5.5 on Oct 7 with tiered pricing that starts at $0.10/$0.50 for prompts up to 100K. Every model in the family carries 1M context, 128K output and half-price batch. This brief closes out the family for API builders who price ladders, not launch events.
What is established: the official List Prices effective 2026-10-07 pin the family — Sonnet 5.5 $2/$10 (cache-hit input 5% = $0.10; batch $1/$5; adaptive thinking, default effort high), Haiku 5.5 tiered at $0.10/$0.50 for prompts ≤100K and $0.50/$2.50 beyond, Opus 5.5 $4/$20. Family-wide: 1M context, 128K max output, June 2026 cutoff, batch 50%. What is not: Anthropic's own savings framing (about 75% cheaper than Haiku 4.5 typically, up to about 90% on short prompts, which it says are roughly 90% of traffic) is a provider claim, not our measurement — and nothing here promises these rates on any relay, this site's included. Buyer takeaway: short-prompt Haiku workloads now price against Chinese budget tiers instead of Western flagships, while Opus remains the 13–17x premium option against DeepSeek's flash line — pick per workload, not per brand.
The facts, as listed (effective Oct 7)
- Sonnet 5.5 (Sep 28): $2/$10, and it sticks. Cache-hit input bills at 5% ($0.10 per 1M) and batch halves to $1/$5. 1M-token context, 128K max output, adaptive thinking on by default at effort high. Anthropic confirmed $2/$10 as the regular price — the Sep 1 notice that would have moved Sonnet to $3/$15 was cancelled, removing the biggest pricing overhang of the quarter.
- Haiku 5.5 (Oct 7): tiered pricing keyed to prompt size. Prompts up to 100K tokens: $0.10 input / $0.50 output per 1M. Beyond 100K: $0.50/$2.50. Official framing puts typical savings at about 75% versus Haiku 4.5, rising to about 90% on short prompts — which Anthropic says make up roughly nine in ten requests.
- Opus 5.5 (Sep 22): $4/$20, already covered. Our dedicated launch brief carries its rate card and cost test; this piece treats it as the family's ceiling and closes the set. Reuters reported the launch; the price holds through the Oct 7 list.
- Family-wide constants. Every 5.5 model lists 1M-token context, 128K max output, a June 2026 knowledge cutoff and 50% batch pricing. For capacity planners that means one context envelope and one discount lever across all three tiers.
- The paper trail. The List Prices PDF effective 2026-10-07 anchors the numbers, alongside the models overview and the Sonnet 5.5 / Haiku 5.5 overview pages on platform.claude.com. Where a launch-day story and the effective list disagree, this brief follows the list.
Rates are provider-published list prices read from Anthropic's pricing pages and the List Prices PDF effective 2026-10-07. No paid API call was made for this article; savings percentages are Anthropic's framing, labeled as such.
Where the family sits against the Chinese ladder
Haiku's short-prompt tier is the interesting one for this site's readers: at $0.10/$0.50 it finally prices a Western frontier-lab model into the neighborhood where Chinese budget tiers live. For comparison, DeepSeek's V4.1-Flash runs $0.30/$1.20 peak with off-peak halves (1017 checkpoint brief) and GLM-5.3-FlashX lists at $0.37/$1.25 — but Haiku bills its cheapest rate only under 100K prompts, while the Chinese rates apply everywhere.
At the top, the gap stays wide: Opus 5.5 at $4/$20 sits roughly 13–17x above DeepSeek's flash line — the multiplier we work through in Hermes Agent on DeepSeek for agent workloads. Sonnet's cancelled increase keeps the middle of the family competitive with Grok 4.7's $2/$6 (tri-model brief).
None of these Western list prices imply relay availability. For Chinese models, the endpoint, model-id and pricing walkthroughs live in our comparison table and per-model guides.
What this means for builders
Tier by prompt shape, not by habit. Short-prompt, high-volume workloads (classification, extraction, routing, chat turns under a few thousand tokens) are what Haiku's $0.10/$0.50 tier was priced for; crossing the 100K prompt boundary quintuples the input rate, so long-context jobs belong on Sonnet unless output dominates the bill.
Two billing behaviors to plan around: Sonnet's thinking runs by default at effort high, so the thinking tokens bill unless you tune it down; and batch's 50% applies across the family for work that tolerates latency — offline evals and summarization pipelines should default there.
And the usual line for this site's readers: these are Anthropic list prices on Anthropic endpoints. A list price is not a routing promise from any relay, including this site's planned service; confirm what your provider actually serves before committing architecture.
Access from outside China — and the usual disclaimer
This site tracks Chinese model APIs for international builders, and prices Western launches only as the comparison frame. The comparison table holds the cross-vendor ladder; per-model guides cover endpoint and model-id detail.
Independence disclaimer: ChinaModelAPI is an independent information site, not affiliated with Anthropic, DeepSeek or any model vendor. Figures are provider-published list prices; savings percentages are the vendor's framing; nothing on this page is a routing or availability promise.
Primary sources
FAQ (2026)
When did the Claude 5.5 family complete?
Opus 5.5 shipped Sep 22, Sonnet 5.5 on Sep 28, and Haiku 5.5 on Oct 7, 2026. The Oct 7 Haiku listing completed the family; its List Prices took effect the same day.
What does Claude Sonnet 5.5 cost?
$2 input / $10 output per 1M tokens, cache-hit input at 5% ($0.10), and batch at $1/$5. 1M-token context with 128K max output; adaptive thinking is on by default at effort high.
What does Claude Haiku 5.5 cost?
Tiered by prompt size: up to 100K tokens costs $0.10/$0.50 per 1M; prompts beyond 100K cost $0.50/$2.50. Same 1M context and 128K output as its siblings.
Is Sonnet 5.5's $2/$10 a promotional price?
No. Anthropic confirmed $2/$10 as the regular price. The Sep 1 plan to raise Sonnet pricing to $3/$15 was cancelled.
How much cheaper is Haiku 5.5 than Haiku 4.5?
Official framing: about 75% cheaper on typical workloads, and up to about 90% cheaper on short prompts — which Anthropic says account for roughly 90% of traffic.
What do all three Claude 5.5 models share?
1M-token context, 128K max output, a June 2026 knowledge cutoff, and 50% off batch pricing. Thinking modes differ per tier; Sonnet defaults to effort high.