← Back to directory
Anthropic
Claude Haiku 4.5
Released Oct 15, 2025 · knowledge cutoff 2025-02
- Status
- Superseded
- Location
- United States
- Modality
- Multimodal
- Context window
- 200K
- Max output
- 64K
- Speed
- —
- Price ($/MTok in / out)
- $1 / $5
- Cost per task
- $0.24
- Open weights
- No
Benchmarks4 of 8 reported
GPQA Diamond
73%
SWE-bench Verified
73.3%
AIME (latest)
83%
LMArena Elo
1405
MMLU-Pro
—
Terminal-Bench 2.1
—
Humanity's Last Exam
—
ARC-AGI-2
—
Strengths & weaknesses
Strengths
- Fast enough for autocomplete, classification and anything where latency is the product
- Small, but can still think before answering — unusual at this size
- Cheap enough to run many copies in parallel as sub-agents under a larger model
Weaknesses
- Loses the thread on long tasks; needs a bigger model to plan for it and check it
- Takes instructions literally and misses implied requirements
Anthropic's current small model.
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor