Claude Sonnet 4.6
Anthropic's February 2026 Sonnet, now a legacy model; Sonnet 5.5 is the current Sonnet.
Anthropic has replaced this model with Claude Sonnet 5.5.
Context window
1M
Input / 1M tokens
$3.00
Output / 1M tokens
$15.00
Provider
Anthropic
Standard pricing $3/$15 per 1M input/output tokens (same as Sonnet 4.5); cache reads $0.30 per 1M. Batch API is 50% off at $1.50/$7.50 per 1M. Now a legacy model: Sonnet 5 and Sonnet 5.5 cost less ($2/$10), though their newer tokenizer counts about 30% more tokens for the same text. · Data verified 2026-11-11
Claude Sonnet 4.6 (February 2026) delivered frontier-class performance across coding, agents, and professional work with a 1M-token context window. It scored 79.6% on SWE-bench Verified and brought strong agentic and computer-use capabilities. Claude Sonnet 5 replaced it in June 2026 at a lower price ($2/$10 vs $3/$15), and Anthropic now lists Sonnet 4.6 as a legacy model and recommends Claude Sonnet 5.5. It stays in the API (retirement not before February 17, 2027) and uses the older tokenizer, which counts about 30% fewer tokens for the same text than Sonnet 5 and later models.
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-sonnet-4-6), Google Cloud, Microsoft Foundry, Claude Platform on AWS, and Amazon Bedrock through the InvokeModel integration (anthropic.claude-sonnet-4-6). Anthropic lists it as a legacy model (retirement not before February 17, 2027) and recommends migrating to Claude Sonnet 5.5.
Strengths
- ✓Strong agentic coding for its tier: 79.6% SWE-bench Verified
- ✓1M-token context window for large inputs
- ✓Solid computer-use and OS-level task performance (OSWorld ~72.5-78.5%)
- ✓Fast latency with good cost-efficiency at $3/$15
- ✓Well-documented, stable, and broadly available across clouds
Best for developers who...
When to choose it (and when not to)
Reach for Claude Sonnet 4.6 when...
- →You have an existing integration pinned to Sonnet 4.6 and want stability
- →You want the older tokenizer, which counts about 30% fewer tokens than Sonnet 5 and later for the same text
- →You still depend on manual thinking budgets, which are deprecated but still accepted on Sonnet 4.6 and rejected on Sonnet 5 and later
- →You need a mid-tier model with 1M context while you test a migration
Look elsewhere if...
- ✕You want the best capability per dollar today: Sonnet 5.5 is stronger and has a lower list price ($2/$10)
- ✕You need the strongest agentic/coding performance (Opus 5.5 or Fable 5.2)
- ✕You want the latest features and long-term support (it is now legacy)
How to use it
- ›Manual extended thinking is deprecated on Sonnet 4.6; prefer adaptive thinking patterns
- ›Use the 1M context deliberately - place key instructions near the top and structure long inputs
- ›Use batch processing for 50% cost savings on non-interactive workloads
- ›For agentic tasks, give explicit tool-use guidance and success criteria
Quickstart
Pythonimport anthropic
client = anthropic.Anthropic()
response = client.messages.create(
model="claude-sonnet-4-6",
max_tokens=4096,
messages=[
{"role": "user", "content": "Summarize this pull request and flag risks."}
],
)
print(response.content[0].text)Sonnet 4.6 is now a legacy model; for new projects migrate to claude-sonnet-5 (drop-in, same price).
API model id: claude-sonnet-4-6
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| SWE-bench Verified | 79.6% | |
| Terminal-Bench 2.0 | 59.1% | |
| OSWorld-Verified | 72.5% | Anthropic later reported 78.5% on an updated run |
| GPQA | 79.7% |
Tools powered by Claude Sonnet 4.6
Compare Claude Sonnet 4.6
Compare Claude Sonnet 4.6 with any other model
Build a comparison →All model comparisons →