← Back to directory
Anthropic
Claude 3.5 Sonnet
Released Jun 20, 2024 · knowledge cutoff 2024-04
- Status
- Deprecated
- Location
- United States
- Modality
- Multimodal
- Context window
- 200K
- Max output
- 8K
- Speed
- —
- Price ($/MTok in / out)
- $3 / $15
- Cost per task
- —
- Open weights
- No
Benchmarks6 of 8 reported
MMLU-Pro
76.1%
GPQA Diamond
59.4%
SWE-bench Verified
49%
AIME (latest)
16%
Humanity's Last Exam
4.8%
LMArena Elo
1300
Terminal-Bench 2.1
—
ARC-AGI-2
—
Strengths & weaknesses
Strengths
- Wrote code that ran on the first try often enough to change how people used LLMs
- Picks up a repo's existing conventions and matches them instead of imposing its own
- Fast enough to sit inside a tight edit-run-fix loop
Weaknesses
- Cuts long files off mid-function, forcing continuation prompts
- Adds unrequested scaffolding — comments, validation, try/except you didn't ask for
- Single-pass reasoning shows immediately on maths and novel puzzles
Scores reflect the upgraded October 2024 snapshot.
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
Retirement date not researched
Lineage
No recorded predecessor