Claude 5.5 Arrives: Opus 5.5 Cuts Prices 20% and Claims 40% Lower Running Cost, Sonnet 5.5 Is Faster at the Same Price, but API Compatibility Needs a Recheck

Anthropic introduced Claude Opus 5.5 (claude-opus-5-5) on September 22, 2026, as the first model in the new Claude 5.5 family. According to Anthropic, Opus 5.5 performs at the level of Claude Fable 5.1 on most work. It is priced at $4 per million input tokens and $20 per million output tokens (20% less than Opus 5), with cache reads at $0.20 (60% less); Anthropic's tests show typical workloads costing 40% less than Opus 5 at default settings, with output more than 30% faster. The release notes list a 1M token context window by default and 128k max output, on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry.

On September 28, Anthropic followed with Claude Sonnet 5.5 (claude-sonnet-5-5), priced the same as Sonnet 5 ($2 input, $10 output, $0.20 cache reads). Anthropic says it is 30%+ faster and up to 30% cheaper per task for most work, and that Claude Haiku 5.5 will join the family in the coming weeks. Anthropic explicitly states that both Opus 5.5 and Sonnet 5.5 are available with zero data retention (ZDR), unlike the mandatory 30-day retention on Fable 5 and Fable 5.1.

Compatibility changes are where upgrades will break. Per the release notes, thinking cannot be disabled on Opus 5.5: both thinking type disabled and enabled return 400, and depth is controlled through effort. The forced tool-use modes tool_choice any and tool also return 400; Anthropic recommends auto with strict tool use. Sonnet 5.5 likewise rejects forced tool use, and its thinking blocks only replay in the account that produced them or a linked account. On safety, Opus 5.5 is the first Opus model with Fable 5.1-class safeguards for cyber, biology, and distillation, with most cybersecurity tasks rerouted to Opus 4.8; preserved thinking, which blocks editing prior context to extract reasoning, applies to API accounts created on or after August 31, 2026.

Fable 5.1-level performance, the 40% cost reduction, and comparisons on Terminal-Bench 4.0, FrontierCode, and GDPval-AA are claims by Anthropic and early-access customers. Savings depend on both lower prices and fewer tokens per task, so real results will vary by workload. Anthropic also acknowledges that Opus 5.5 often suspects it is being evaluated, which limits confidence in pre-release assessments.

Before moving to the Claude 5.5 family, check each item: remove all thinking fields from requests (both disabled and enabled return 400 on Opus 5.5) and control depth with effort; on Sonnet 5.5, turning off up-front thinking means sending thinking type "between_tools", only at high effort or below; change tool_choice any and tool to auto, and use strict tool use or structured outputs where a fixed format is required; for computer use on the Claude API and Google Cloud, the older computer_20251124 tool is no longer accepted, so move to computer_toolset_20260801; expect the API to silently drop Sonnet 5.5 thinking blocks replayed from another account; and, as Anthropic recommends, migrate integrations still on claude-sonnet-4-5-20250929 before its November 30 retirement.

Things to watch: Haiku 5.5 pricing and timing, the tiers of the Cyber Verification Program once it expands to Opus 5.5, and the retirement of Claude Sonnet 4.5 on November 30, 2026, announced on September 30, which puts integrations still on older models onto a migration clock.