Claude Opus 5.5 and Sonnet 5.5: API Pricing, Migration Changes, and Model Choice
Three Model Changes in September: The Claude Lineup Now
In September 2026, three positions in Anthropic's current lineup changed: on September 1, Fable 5.1 succeeded Claude Fable 5; on September 22, Opus 5.5 launched; on September 28, Sonnet 5.5 launched. Haiku 5.5 followed on October 7, so all four current positions are now on the new generation.
The change buyers should notice most is that the official default recommendation has changed. Anthropic's models overview now says: if you're unsure which model to use, start with Claude Opus 5.5 for most workloads; use Fable 5.1 for demanding reasoning and long-horizon agentic work, or when Opus 5.5 at higher effort still falls short.
Current Lineup and Pricing
| Model | Model ID | Input/Output (per million tokens) | Cache read | Context | Max output | Official positioning |
|---|---|---|---|---|---|---|
| Claude Fable 5.1 | claude-fable-5-1 | $10/$50 | $0.25 | 1M | 128K | Demanding reasoning and long-horizon agents |
| Claude Opus 5.5 | claude-opus-5-5 | $4/$20 | $0.20 | 1M | 128K | Long-running agentic coding and knowledge work |
| Claude Sonnet 5.5 | claude-sonnet-5-5 | $2/$10 | $0.10 | 1M | 128K | Best combination of speed and intelligence |
| Claude Haiku 5.5 | See the official Haiku 5.5 page | $0.10/$0.50 for prompts up to 100K tokens; $0.50/$2.50 above | $0.01 (up to 100K) / $0.05 | 1M | 128K | Lowest price in the current lineup |
Sources: Anthropic's official models overview and pricing pages (checked 2026-09-30; Haiku 5.5 and the Sonnet 5.5 cache-read price rechecked 2026-10-28), USD. Fable 5, Opus 5, Sonnet 5, Opus 4.8/4.7/4.6/4.5, Sonnet 4.6, and Haiku 4.5 ($1/$5, now succeeded by Haiku 5.5) are legacy (still available); Sonnet 4.5 was deprecated on 2026-09-30 and retires on 2026-11-30, with Sonnet 5.5 as Anthropic's recommended replacement.
Claude Opus 5.5: A Cheaper New Default
Pricing: Lower Than Opus 5 on Every Line
| Item | Opus 5.5 | Opus 5 |
|---|---|---|
| Input | $4 | $5 |
| Output | $20 | $25 |
| 5-minute cache write | $5 | $6.25 |
| 1-hour cache write | $8 | $10 |
| Cache read | $0.20 | $0.50 |
| Batch API (input/output) | $2/$10 | $2.50/$12.50 |
The biggest gap is the cache read: on Opus 5.5 it is 5% of the input price (also 5% on Sonnet 5.5; 10% on most models), so the more often you resend the same system prompt, the more you save.
Key Specs
- 1M-token context and up to 128K output tokens per request (up to 300K on the Batch API with a beta header)
- Adaptive thinking is always on and cannot be turned off; control thinking depth with the effort parameter instead. The default effort on the Claude API is
medium - Fast mode (research preview) is available on the Claude API at $8/$40
Claude Sonnet 5.5: An Upgrade at the Same Price
Sonnet 5.5 is priced the same as Sonnet 5 ($2/$10); its cache reads are 5% of the input price ($0.10), below Sonnet 5's $0.20. It is positioned as the best combination of speed and intelligence and a step up from Sonnet 5. It also has a 1M context and 128K output; adaptive thinking is on by default, but the lowest setting, between_tools, turns off up-front thinking (at high effort or below). The default effort on the Claude API is high.
Sonnet 5 was originally launched with introductory pricing through the end of August; Anthropic has confirmed that $2/$10 is now the standard price and that the previously scheduled increase to $3/$15 on September 1 will not occur, so Sonnet 5 and Sonnet 5.5 currently cost the same.
Cost Math (Arithmetic, Not Benchmarks)
Assume 50 million input tokens and 10 million output tokens per month, no caching:
| Model | Input cost | Output cost | Monthly total |
|---|---|---|---|
| Fable 5.1 | $500 | $500 | $1,000 |
| Opus 5.5 | $200 | $200 | $400 |
| (Reference) Opus 5 | $250 | $250 | $500 |
| Sonnet 5.5 | $100 | $100 | $200 |
| Haiku 5.5 (each prompt up to 100K tokens) | $5 | $5 | $10 |
| (Reference) Haiku 4.5 | $50 | $50 | $100 |
If Haiku 5.5 prompts exceed 100K tokens, the rate becomes $0.50/$2.50, so the same volume costs $25 + $25 = $50. Haiku 5.5 also uses a new tokenizer that produces about 30% more tokens than Haiku 4.5 for the same text, so the real gap is smaller than the table shows.
If 40 million of the 50 million input tokens hit the cache, Opus 5.5 input costs $8 (cached) + $40 (uncached) = $48, versus $20 + $50 = $70 on Opus 5. This simplified math ignores cache-write costs, which are billed separately at 1.25x (5-minute) or 2x (1-hour) the input price.
⚠️ Watch token counts when comparing across generations: Anthropic notes that Claude 4.7 and later models use a newer tokenizer that produces roughly 30% more tokens for the same text (depending on content). If you're moving from Sonnet 4.6 or earlier, don't compare unit prices alone.
Migrating from Opus 5 or Sonnet 5: Changes That Return 400
Of the changes listed in the official release notes and migration guides, these will make existing code fail outright:
- Forced tool use is no longer supported:
tool_choiceofanyortoolreturns 400 (on both Opus 5.5 and Sonnet 5.5). Useautowith strict tool use instead - Thinking can't be disabled on Opus 5.5: sending
thinking: {"type": "disabled"}or{"type": "enabled", ...}returns 400 — omit thethinkingfield and use effort - To turn off up-front thinking on Sonnet 5.5: send
thinking: {"type": "between_tools"}instead of"disabled" - Computer use tool version change: on the Claude API and Google Cloud, the earlier
computer_20251124is no longer accepted; usecomputer_toolset_20260801. On Amazon Bedrock the older version still works
Anthropic also lists "thinking blocks are tied to the model and the conversation" as a change; in addition, thinking blocks produced by Sonnet 5.5 work only in the account that produced them (or a linked account), and when another account sends one, the API drops the block and the request still succeeds. Applications that replay thinking content should be retested.
Sampling parameters are not a change introduced by this generation: non-default temperature, top_p, or top_k returns 400 on both Opus 5.5 and Sonnet 5.5, but Anthropic states this applies to every model from Opus 4.7 onward, so Opus 5 and Sonnet 5 already return 400. You will only run into it for the first time when moving from an earlier model.
CloudInsight Enterprise Claude API Procurement | Local Invoices, No Foreign Credit Card Needed
With model generations changing this fast, what enterprises get stuck on most is not the technology but payment, invoicing, and multi-platform billing.
Contact CloudInsight for a Claude API Enterprise Quote
We offer one-stop procurement for Claude, GPT, and Gemini, enterprise discounts, Taiwan uniform invoices, and technical support in Chinese.
How to Choose
| Workload | Recommendation | Why |
|---|---|---|
| Unsure — want to ship first | Opus 5.5 | Anthropic's stated default |
| High-volume support, content rewriting, small code fixes | Sonnet 5.5 | Half the price of Opus 5.5 and faster |
| Classification, routing, real-time responses | Haiku 5.5 | Cheapest in the current lineup ($0.10/$0.50 for prompts up to 100K tokens, $0.50/$2.50 above), 1M context |
| Hardest reasoning, research-style agent tasks | Fable 5.1 | Upgrade when Opus 5.5 at higher effort still falls short |
Haiku 4.5 has been succeeded by Haiku 5.5 (launched 2026-10-07) and is now legacy, though still available. Its retirement commitment of "not sooner than October 15, 2026" has passed without a retirement date being announced, so long-running projects should move to Haiku 5.5 (retirement not sooner than 2027-10-07).
FAQ: Claude Opus 5.5 and Sonnet 5.5
How much does Claude Opus 5.5 cost?
$4 per million input tokens and $20 per million output tokens, $0.20 for cache reads, and half price on the Batch API ($2/$10).
Is Opus 5.5 cheaper than Opus 5?
Yes. Opus 5 is $5/$25; Opus 5.5 is lower on input, output, cache writes, and cache reads, with input and output each 20% cheaper.
When did Claude Sonnet 5.5 launch, and did the price go up?
It launched on 2026-09-28 at the same price as Sonnet 5: $2/$10.
Can we still use Opus 5 and Sonnet 5?
Yes. Anthropic lists them as legacy (still available), with retirement commitments of "not sooner than 2027-07-24" for Opus 5 and "not sooner than 2027-06-30" for Sonnet 5. New projects should still start on the 5.5 generation.
What code changes does Opus 5.5 need?
The three most common: remove thinking disabled/enabled settings, change tool_choice from any/tool to auto, and move computer use to the new toolset.
Conclusion: Default to Opus 5.5, Push Volume Work Down to Sonnet 5.5
Three Core Recommendations
- Default new projects to Opus 5.5: 20% cheaper than Opus 5 and Anthropic's recommended starting point
- Always cache long prompts you resend: Opus 5.5 cache reads cost just 5% of the input price
- Run a test pass before migrating:
tool_choice,thinking, and sampling parameters are the settings most likely to return 400
Ready to Adopt Claude 5.5?
Contact the CloudInsight sales team and we'll help plan your Opus 5.5 and Sonnet 5.5 usage mix and handle payment and invoicing.
Join our LINE Official Account for instant model-selection consultation.
Further Reading
- Claude Haiku 5.5 Guide (the new Haiku released 2026-10-07: pricing and migration details)
- Claude Fable 5 Complete Guide
- Claude API Purchase Guide
- Claude API Pricing Explained
- GPT-6 Astra, 6.1 Sol, Luna: Which to Pick
References
- Models overview — Claude Platform Docs (checked 2026-09-30)
- Pricing — Claude Platform Docs (checked 2026-09-30)
- Claude Opus 5.5 — Claude Platform Docs
- Claude Sonnet 5.5 — Claude Platform Docs
- Claude Haiku 5.5 — Claude Platform Docs (checked 2026-10-28)
- Release notes — Claude Platform Docs
- Model deprecations — Claude Platform Docs
- Claude Opus 5.5 is now available on AWS — AWS What's New (2026-09-22)
- Claude Sonnet 5.5 now available on AWS — AWS What's New (2026-09-28)
Need Professional Cloud Advice?
Whether you're evaluating cloud platforms, optimizing existing architecture, or looking for cost-saving solutions, we can help
Book Free ConsultationRelated Articles
Claude Haiku 5.5 Guide: Tiered API Pricing, Differences from Haiku 4.5, and Migration
Anthropic released Claude Haiku 5.5 on 2026-10-07. For prompts up to 100,000 tokens it costs just $0.10/$0.50 per million tokens, one tenth of Haiku 4.5. But a request whose prompt is over 100,000 tokens is billed at $0.50/$2.50 in full, and the new tokenizer counts the same text as roughly 30% more tokens. This guide covers official specs, both price tiers, cost math that accounts for both traps, and the changes that break or alter behavior when migrating from Haiku 4.5.
AI APIGemini 4 Argon Guide: API Pricing (Intro Rate Through Nov 30), Specs, and Migrating from Gemini 3
Google made its flagship Gemini 4 Argon generally available (GA) to developers on 2026-10-13. The introductory price of $2/$10 only runs through 2026-11-30; from 12-01 it moves to $4/$20, and there is no free tier. This guide covers the official specs (1M context, output up to 1M, three thinking levels), both price stages for every consumption option, cost math, how it splits work with Gemini 3.8 Flash, and the settings to change when moving from the Gemini 3 family.
AI APIClaude Fable 5 Complete Guide 2026: The First Mythos-Tier Model — Features, Benchmarks & Enterprise Procurement
In June 2026 Anthropic released Claude Fable 5, the first publicly available Mythos-tier model. It tops SWE-Bench Pro at 80.3%, costs exactly double Opus 4.8 ($10/$50 per million tokens), and landed on AWS Bedrock and Google Cloud on launch day. This guide covers features, benchmarks, pricing, and procurement paths for Taiwanese enterprises.