← leaderboard
Claude Haiku 4.5
NOT COOKEDAnthropic · anthropic/claude-haiku-4.5 · verdict as of 2026-08-10
Daily test history (90 days, baseline band = trailing mean ± 2σ)
Code Generation
NOT COOKED
Math & Reasoning
NOT COOKED
Game Design
NOT COOKED
Instruction Following
NOT COOKED
Report Analysis
NOT COOKED
Summarization
NOT COOKED
Customer Service
NOT COOKED
Web Design
NOT COOKED
Creative Writing
no data
Structured Extraction
NOT COOKED
Serving providers (last 30 days)
striped = multiple providers served this model that day (hover for detail)
Public benchmarks overall 58.6
MMLU-Pro
—
GPQA Diamond
78.2
SWE-bench Verified
66.6
LMArena Elo
1411
AIME 2025
—
retrieved 2026-07-03 from public sources — see methodology
Recent samples (latest run, one per test case)
No samples from the latest run.