Opus 5.5 is available in Claude Code from launch day, and Anthropic’s launch numbers explain why people want it there: 66.4% on Terminal-Bench 4.0 and 57.8% on CursorBench 4.0, both ahead of Fable 5.1. Claude Code speaks the Anthropic Messages API, so any gateway that serves /v1/messages can run it. apimodels.app does, with prompt caching passed through.
Claude Code sends the whole working context on every turn, so input tokens dominate and prompt caching matters. Opus 5.5 cache reads list at $0.20 per million and cache writes at $5; on apimodels.app they are $0.12 and $3.00. As a rough guide, a turn that re-sends 60,000 cached tokens and writes 2,000 tokens of code costs about $0.007 for the cached input plus $0.024 for the output on apimodels.app.
| Per 1M tokens | Anthropic list | apimodels.app |
|---|---|---|
| Input | $4 | $2.40 |
| Output | $20 | $12 |
| Cache read | $0.20 | $0.12 |
| Cache write | $5 | $3.00 |
Thinking is always on for Opus 5.5; you control depth with effort (low, medium, high, xhigh, max), and the default is medium. Medium is right for most edits; raise it for large refactors or debugging you have already failed at once. Reasoning tokens bill as output, so higher effort costs more per turn.
Coming from Opus 5: four things break. Thinking cannot be disabled, forcing a specific tool with tool_choice returns an error, thinking blocks are bound to the model that produced them, and the old computer_20251124 tool is no longer accepted. Claude Code itself handles all four; custom agents built on the API need to check them.
macOS / Linux (~/.zshrc or ~/.bashrc)
export ANTHROPIC_BASE_URL="https://api.apimodels.app"
export ANTHROPIC_AUTH_TOKEN="sk_…your_apimodels_key…"
export ANTHROPIC_MODEL="claude-opus-5-5"
# then: source ~/.zshrc && claudeWindows (PowerShell)
[Environment]::SetEnvironmentVariable("ANTHROPIC_BASE_URL", "https://api.apimodels.app", "User")
[Environment]::SetEnvironmentVariable("ANTHROPIC_AUTH_TOKEN", "sk_…your_apimodels_key…", "User")
[Environment]::SetEnvironmentVariable("ANTHROPIC_MODEL", "claude-opus-5-5", "User")
# open a new terminal, then run: claudecURL
curl https://api.apimodels.app/v1/messages \
-H "x-api-key: $APIMODELS_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-5-5",
"max_tokens": 16000,
"messages": [{"role": "user", "content": "Review this function for race conditions: ..."}]
}'Yes. Set ANTHROPIC_BASE_URL to https://api.apimodels.app, put your apimodels key in ANTHROPIC_AUTH_TOKEN and set ANTHROPIC_MODEL to claude-opus-5-5. You pay per token, $2.40 / $12 per million, with no monthly fee.
Not through any API we know of: Anthropic lists it at $4 / $20 per million tokens and resellers charge per token too. The cheapest way to use it occasionally is per-token billing with no subscription; on apimodels.app a short coding turn usually costs a few cents. Pages promising free unlimited Opus 5.5 access are worth treating with suspicion.
Yes. Cache-control markers are passed through to the upstream, and cache reads bill at $0.12 per million, a twentieth of the input price. Claude Code sets them automatically.
Anthropic reports Sonnet 5.5 at 70.6% on Terminal-Bench 4.0 against Opus 5.5’s 66.4%, at half the list price, so for well-scoped edits Sonnet 5.5 is a strong default. Opus 5.5 is the better choice for large, ambiguous changes. Sonnet 5.5 is not on apimodels.app yet; Sonnet 5 is.
