Claude Opus 4.6 API - pricing, context window and best use cases

Claude Opus 4.6 is Anthropic's high-end reasoning model for hard, long-horizon work: large refactors, deep research, agentic loops, and document-heavy analysis. This guide covers what actually matters when you put it into production - pricing, the context window, the model id - and where it earns its cost versus a cheaper model.

Pricing

Opus 4.6 is priced per million tokens (MTok), split into input and output:

Two practical notes. First, the 1M-token context window is served at standard pricing - there is no long-context premium tacked on once you cross 200K. Second, output tokens cost 5x input, so the biggest lever on your bill is how much the model writes, not how much you send. Prompt caching helps a lot here: cached input is read at roughly a tenth of the base rate, so a large, stable system prompt or document prefix gets cheap on every request after the first.

If you want to sanity-check Opus 4.6 against Sonnet or a non-Anthropic model before committing, the model comparison page lays the tiers out side by side.

Context window and output limits

• Context window: 1,000,000 tokens (input + output combined).
• Max output: 128,000 tokens per response.

A million tokens is enough to drop an entire mid-size codebase, a long PDF corpus, or a multi-hour conversation history into a single request. For outputs above ~16K tokens you should stream the response - non-streaming requests that large risk HTTP timeouts on most SDKs. The 128K output ceiling means Opus 4.6 can produce a full report or a large generated file in one turn rather than forcing you to paginate.

Model id and how to call it

The model id is claude-opus-4-6. On AnyModel you don't manage a separate Anthropic account or SDK - it's one OpenAI-compatible endpoint where a single API key reaches every model (GPT, Claude, Gemini, DeepSeek, GLM, Kimi, Qwen, Grok). Switching models is just changing the model field; nothing else in your code changes.

Base URL: https://anymodel.org/v1

curl https://anymodel.org/v1/chat/completions \
  -H "Authorization: Bearer $YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-4-6",
    "messages": [{"role": "user", "content": "Summarize this repo and propose a refactor plan."}]
  }'

For terminal tools, there's a one-line install (supported tools: codex, claude, opencode, hermes):

bash <(curl -fsSL "https://anymodel.org/i?tool=claude") <YOUR_API_KEY>

For Cursor, Windsurf, Zed, Cline, Aider, Continue, or Gemini CLI, use a manual OpenAI-compatible setup: point the base URL at https://anymodel.org/v1 and paste your key. The full id list lives on the models page.

Adaptive thinking and effort

Opus 4.6 uses adaptive thinking - the model decides when and how much to reason, rather than you setting a fixed token budget. You tune the trade-off with an effort level (low, medium, high, or max). high is the usual sweet spot; drop to medium for cost-sensitive routes, push to max when correctness matters more than spend. There's no budget_tokens knob to manage anymore.

Best use cases

Opus 4.6 is worth its price when the task is genuinely hard:

• Agentic coding - multi-file refactors, debugging, repo-wide changes that need the model to hold a lot of context and self-correct.
• Deep research and synthesis - feed it a large document set and let the 1M window do the work in one pass.
• Long-horizon autonomous loops - tool-using agents that run many steps without losing the thread.
• Knowledge work on documents - spreadsheets, slides, and reports where the model verifies its own output.

When the job is high-volume classification, summarization, or simple chat, a cheaper tier usually wins on cost-per-result - reserve Opus 4.6 for the work that justifies it.

Privacy

If your prompts are sensitive, Ghost Mode offers opt-in zero-retention API keys: AnyModel doesn't store prompts or responses, only a token counter. Note the model provider still receives the prompt to generate a reply - this isn't "100% privacy," but it removes our side from the retention picture.

Get started

You can try Opus 4.6 free: every signup gets 1,000,000 tokens, rising to 6,000,000 total if you link Telegram - no credit card. After that it's pay-per-token, with no subscription and no minimums. More walkthroughs are on the blog.

Create your free AnyModel account and call claude-opus-4-6 in minutes.