← Back to directory
Anthropic
Claude Sonnet 4.5
Released Sep 29, 2025 · knowledge cutoff 2025-07
- Status
- Superseded
- Location
- United States
- Type
- LLM
- Modality
- Multimodal
- Context window
- 200K
- Max output
- 64K
- Speed
- —
- Price ($/MTok in / out)
- $3 / $15
- Cost per task
- $0.41
- Open weights
- No
Benchmarks7 of 10 reported
GPQA Diamond
83.4%
SWE-bench Verified
77.2%
SWE-bench Pro
43.6%
AIME (latest)retired
87%
Humanity's Last Exam
17.3%
LMArena Elo
1449
ARC-AGI-2
13.6%
MMLU-Proretired
—
Terminal-Bench 2.1
—
GDPval-AA v2
—
Strengths & weaknesses
Strengths
- Sustains day-long autonomous runs without drifting off the original goal
- Drives a real desktop — clicks, scrolls and reads the screen — reliably enough to trust with a workflow
- Manages its own context, summarising and discarding what it no longer needs
Weaknesses
- Hits a ceiling on genuinely novel reasoning; more thinking doesn't produce a better answer
- Declares a task finished before verifying that it is
SWE-bench Pro 43.6% is Scale's public-set figure (also Table 1 of the SWE-Bench Pro paper, arXiv 2509.16941) — third-party, not Anthropic-reported.
For developers
API model strings
- anthropic
claude-sonnet-4-5-20250929 - aws-bedrock
anthropic.claude-sonnet-4-5-20250929-v1:0 - google-vertex
claude-sonnet-4-5@20250929
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor