Claude Opus 4.7
The April 2026 Opus with improved high-resolution vision, now a legacy model; Opus 5.5 is the current Opus.
Anthropic has replaced this model with Claude Opus 5.5.
Context window
1M
Input / 1M tokens
$5.00
Output / 1M tokens
$25.00
Provider
Anthropic
Rate card unchanged from Opus 4.6 at $5/$25 per 1M input/output tokens, but Opus 4.7 shipped a new tokenizer that can produce roughly 1.0-1.35x more tokens for the same text, so real per-request cost can rise. Up to 90% savings with prompt caching, 50% with batch. Fast mode is no longer offered on Opus 4.7. Now a legacy model: Opus 5.5 costs less at $4/$20. · Data verified 2026-11-08
Claude Opus 4.7 (April 2026) advanced agentic coding and added substantially better vision, processing images up to ~3.75 megapixels (over 3x prior Claude models). It scored 87.6% on SWE-bench Verified and introduced adaptive thinking only (no manual thinking budgets) plus a new tokenizer. Claude Opus 4.8 replaced it in May 2026, and Anthropic now lists Opus 4.7 as a legacy model and recommends migrating to Claude Opus 5.5 ($4/$20). It stays in the API with retirement not before April 16, 2027.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Available via the Claude API (claude-opus-4-7), Amazon Bedrock (anthropic.claude-opus-4-7), Google Cloud, Microsoft Foundry and Claude Platform on AWS. Anthropic lists it as a legacy model (retirement not before April 16, 2027) and recommends migrating to Claude Opus 5.5.
Strengths
- ✓Top-tier agentic coding: 87.6% SWE-bench Verified
- ✓Much-improved high-resolution vision (up to ~3.75 MP images)
- ✓1M-token context with 128k max output
- ✓Solved coding tasks neither Opus 4.6 nor Sonnet 4.6 could
- ✓Adaptive thinking with effort control
Best for developers who...
When to choose it (and when not to)
Reach for Claude Opus 4.7 when...
- →You have a pipeline pinned to Opus 4.7 and need stability
- →Your workload is vision-heavy and already tuned to Opus 4.7's high-resolution image input
- →You rely on thinking type disabled or forced tool_choice, which Opus 5.5 rejects
- →You want documented, broadly available Opus-class behavior while you test newer models
Look elsewhere if...
- ✕You want the current Opus: Opus 5.5 is stronger and cheaper ($4/$20)
- ✕You are cost-sensitive - note the tokenizer can inflate real per-request cost
- ✕You need manual extended-thinking budgets or custom sampling params (400 error)
How to use it
- ›Recount token budgets - the tokenizer can yield 1.0-1.35x more tokens than earlier Opus models
- ›Use adaptive thinking with the effort parameter; manual budgets return 400
- ›Feed higher-resolution images to exploit the improved vision pipeline
- ›Remove non-default temperature/top_p/top_k (returns 400)
Quickstart
Pythonimport anthropic
client = anthropic.Anthropic()
response = client.messages.create(
model="claude-opus-4-7",
max_tokens=8000,
thinking={"type": "adaptive"},
output_config={"effort": "high"},
messages=[
{"role": "user", "content": "Review this diagram and generate the API spec."}
],
)
print(response.content[0].text)Legacy model - for new work migrate to claude-opus-4-8. Manual thinking budgets and non-default sampling params return 400.
API model id: claude-opus-4-7
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| SWE-bench Verified | 87.6% | |
| SWE-bench Pro | 64.3% | Per comparison tables vs Opus 4.8's 69.2% |
Source: MorphLLM Claude benchmark table
Tools powered by Claude Opus 4.7
Compare Claude Opus 4.7
Compare Claude Opus 4.7 with any other model
Build a comparison →All model comparisons →