Anthropic · The Agentic Era
Claude Opus 4.7
Anthropic's flagship at the time of this entry: extended thinking on, 1M-token context, the most patient of the frontier models.
By C.W. Jameson · Published 19 May 2026 · Last reviewed 19 May 2026
Opus 4.7 is the model Anthropic ships to operators who care more about being correct than being quick. It will take a paragraph to settle on an answer, run twelve tool calls before replying, and refuse a prompt that an older Claude would have politely attempted. On long-horizon coding tasks (the ones requiring an hour of effort and a hundred file reads), it is the only commercial model that finishes them without operator hand-holding.
Field signature
Will explicitly write 'I notice…' or 'Let me check…' before any non-trivial tool call. Extended-thinking blocks are visibly literate.
Specifications
| Released | 2026 |
|---|---|
| Context window | 1,000,000 tokens (200K standard, 1M opt-in) |
| Pricing | $15 / $75 per million tokens (input / output, May 2026) |
| Modalities | text · image · PDF |
| License | Commercial API only, no weights |
| Era | The Agentic Era |
Strengths
- Multi-hour autonomous coding sessions without losing context or drifting
- Refuses politely and specifically rather than vaguely
- Tool-use planning paragraphs read like a research engineer's notebook
Weaknesses
- Slower than Sonnet 4.6 on simple lookups
- Output token cost punitive for high-volume generation
- Extended thinking inflates billed-output 3-5×
Authentication markers
The fingerprints by which Claude Opus 4.7 can be identified from its output alone.
| Tell | Meaning |
|---|---|
| Phrases 'I'll note that…' before tool calls. | Opus 4.x lineage. Other models call tools without commentary. |
| Refuses with a specific alternative ('I can do X but not Y because Z') rather than 'I cannot help with that'. | Constitutional-AI training surfacing. |
| Will say 'measure twice, cut once' or 'check with the user before proceeding' verbatim in the Claude-Code harness. | Anthropic system-prompt artefact. |
Notable works
- Powering Claude Code's long-horizon refactor loops
- Default backbone for serious agent harnesses that need 1M-token context
Market position
$15-$75 per million tokens, May 2026
Partner offer
Anthropic's Claude family is the model lineage most operators end up on for serious agent work. The free tier remains useful.
Try Claude →Affiliate link — see disclosure.
Primary sources
From the Almanac shop
The Operator's Compendium
Every agent harness, every routing pattern, every cost trick. 90-page PDF.
$29 — Coming soon