Claude Opus 5.5 API
NewLLMThe Claude Opus 5.5 API is Anthropic's newest Opus route: the September 2026 rate card drops to $4 / $20 per million tokens while keeping 1M context, vision, tool use, and always-on adaptive thinking.
- Input · USD / 1M tokens$1.12 USDOfficial price
$4 USD - Output · USD / 1M tokens$5.6 USDOfficial price
$20 USD - Savings72%
- Context1M
- View tool pricing and billing rules
What are we making today?
Bring an idea, a question, or a first draft.
01
What this model is for
The Claude Opus 5.5 API is Anthropic's newest Opus route: the September 2026 rate card drops to $4 / $20 per million tokens while keeping 1M context, vision, tool use, and always-on adaptive thinking.
Claude Opus 5.5 is Anthropic's Opus release of September 22, 2026. It keeps the 1M-token context window and 128K standard output limit of Opus 5, takes text and images, and returns text. The catalog lists vision, tool use and adaptive thinking for it, and thinking cannot be switched off. Within the Claude line it sits above Sonnet 5.5 in cost and is built for long, judgment-heavy engineering work rather than quick edits.
Reach for Opus 5.5 when the expensive part of a job is deciding what is wrong: failures that cross services, migrations with quiet compatibility rules, reviews where a second pass costs more than the first. Its default API effort is medium, unlike the high default of Opus 5 and Sonnet 5.5, so start there and raise effort only when your own test still fails. Anthropic's model guide starts most work on Opus 5.5; teams that count the cost of each finished change often open on Sonnet 5.5 instead.
Best-fit coding workloads
Cross-service failure triage
Hand over the failing request, the logs from each affected service and the deploy timeline, then ask for a ranked list of causes and the signal that would confirm each one. The 1M window holds several services at once. Require the confirming check before any patch is written.
Migrations that keep old clients working
Plan a schema or endpoint migration where existing callers must keep running. Supply the current interface, the call sites and the compatibility notes, and ask for ordered stages with a rollback step for each. Keep the opening prompt identical across turns so cache reads stay valid.
Screenshot review of shipped UI
Send screenshots of a page alongside the component source and ask for visual defects with file references. Image input is listed for this model, so pictures and code can travel in one request. Crop to the region under review; a full-page shot of a dense dashboard loses readable detail.
02
Hands-on notes
Checked on 2026-10-27 through the OmniaKey API with one standard coding prompt (median of 3 runs) and three capability probes.
- Output tokens for our standard coding prompt, reasoning included
- 642
- Cost of that output at current prices
- $0.0036
- JSON mode
- Passed
- Tool calling
- Passed
- Image input
- Passed
03
Technical specification
Catalog facts and integration details verified for this route.
- Model ID
- claude-opus-5-5
- Context
- 1M
- Inputs and tools
- Vision · Tool use · Adaptive thinking
- API endpoint
- /v1/chat/completions
- Cache hit / 1M
- $0.056 USD
- Coding clients
- Claude Code · Cursor · Cline · Aider
Limits to plan around
- Thinking cannot be turned off. A request that sends thinking disabled, or enabled with a manual token budget, is rejected; lower output_config.effort when you need less cost or less waiting, then re-run your test.
- Forced tool choice is no longer accepted: tool_choice types any and tool return an error. Workflows built on them must ask for the tool in the instruction and validate arguments with a strict schema instead.
- Text between tool calls can arrive in thinking blocks that the default display setting leaves empty, so a client that renders only text blocks looks stalled while the request is still running.
- The 1M window is capacity, not recall. A large paste can bury the two files that matter, and the 300K output ceiling belongs to a beta batch path rather than ordinary requests.
04
Compare nearby models
Compare workload fit and current listed token rates before choosing a route.
| Model | Best for | Context | Input | Output |
|---|---|---|---|---|
| Claude Opus 5.5This page claude-opus-5-5 | Hard, ambiguous, expensive-to-get-wrong engineering work where a wrong early assumption costs more than the tokens. | 1M | $1.12 USD Official price $4 USD | $5.6 USD Official price $20 USD |
| Claude Sonnet 5.5 claude-sonnet-5-5 | Daily coding and document work with tests that can reject a weak patch, at lower input and output rates. | 1M | $0.56 USD Official price $2 USD | $2.8 USD Official price $10 USD |
| Claude Opus 5 claude-opus-5 | Agents still pinned to the older Opus 5 ID, or a staged switch that has not replayed its prompts yet. | 1M | $1.4 USD Official price $5 USD | $7 USD Official price $25 USD |
| Claude Fable 5.1 claude-fable-5-1 | The longest autonomous sessions where an Opus 5.5 run at higher effort still misses the goal. | 1M | $2.8 USD Official price $10 USD | $14 USD Official price $50 USD |
05
SDK quickstarts
Use one base URL and model ID from cURL, Python, or TypeScript.
curl https://api.omniakey.com/v1/chat/completions \
-H "Authorization: Bearer $OMNIAKEY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-5-5",
"messages": [
{"role": "user", "content": "Review this pull request"}
],
"stream": true
}'06
Estimate token cost
Enter expected input and output tokens to compare current listed rates.
- OmniaKey estimate
- $0.224
- Official-rate estimate
- $0.80
- Estimated savings
- $0.576
Estimate covers input and output tokens only. Cache usage, retries, tools, and client-side token accounting can change the final charge.
07
Related coding guides
Deeper selection, setup, and troubleshooting guidance for this model family.
Claude Opus 5.5 review
Check Opus 5.5’s release, pricing and migration requirements.
Read guideOpus 5.5 vs Sonnet 5.5
Choose by results and total cost.
Read guideBest model for Claude Code
Start daily coding with Sonnet 5.5 at $2 per million input tokens, move complex multi-file work to Opus 5.5 at $4, keep Fable 5.1 for the hardest long sessions, and hand quick edits and subagents to Haiku 5.5.
Read guideOpus 5.5 video guide
Turn a clear brief into editable motion graphics and an MP4.
Read guide08
Sources and verification
Model facts and links were checked on 2026-10-27. Provider documentation remains the authority for upstream behavior.
Frequently asked questions
How much does Claude Opus 5.5 cost on OmniaKey?
OmniaKey charges $1.12 USD per 1M input tokens and $5.6 USD per 1M output tokens, in USD. Official prices are $4 USD and $20 USD; you save 72%.
What is the context window for Claude Opus 5.5?
The listed context window is 1M. Actual usable limits can also depend on the client and upstream request format.
Is Claude Opus 5.5 a better choice than GPT-6.1 Sol for coding?
They answer different constraints. Sol is the lower-cost opening when your stack already speaks OpenAI's Responses API. Opus 5.5 suits long investigations inside Anthropic Messages and Claude Code, where thinking stays on and effort is yours to set. The two vendors published coding results under different harnesses, so settle the choice with tasks from your own repository.
How is Claude Opus 5.5 billed?
It is metered per token: uncached input, output (which includes billed thinking), cache reads, and cache writes at either the five-minute or the one-hour lifetime. There is no prompt-length step for this model, so one rate card covers the whole 1M window. The current rates appear in the price table beside the model on the catalog.
What model id and endpoint do I use on OmniaKey?
Use claude-opus-5-5 with POST /v1/chat/completions on https://api.omniakey.com/v1 and the model id in the model field. For Claude Code, set ANTHROPIC_BASE_URL to https://api.omniakey.com and ANTHROPIC_AUTH_TOKEN to your OmniaKey key, then select the same id. Model ids are case-sensitive.
Can I disable thinking on Claude Opus 5.5?
No. Adaptive thinking is always on for this model, and the old disabled form, or an enabled form with budget_tokens, comes back as an error. Control depth with output_config.effort instead, and leave enough max_tokens for both the thinking and the visible answer. Anthropic's migration guide covers the remaining request changes.
How much does the context window cost me?
Nothing by itself: the 1M window is a capacity limit, and only tokens you actually send or receive are billed. Output above the 128K standard ceiling is not available on normal requests. Long prompts are cheaper when the opening part repeats, because a reported cache read is charged at the cache-read rate rather than the input rate.
Can I run Claude Opus 5.5 in Claude Code?
Yes, once Claude Code is at version 2.1.280 or later; run claude update if the model is missing from the picker. Then use /model claude-opus-5-5. Anthropic's release notes also list availability through the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS, each with its own terms.