← Claude Sonnet 5 · API · High effort (default)
Evidenceacceptedthird_party_leaderboard
Artificial Analysis API latency / throughput
Artificial Analysis — Claude Sonnet 5 (high) · Anthropic provider
ev_2026_07_sonnet5_aa_speed
Result
71
Confidence B
- Tested
- 2026-07-01
- Runs
- 1
Configuration
How it was measured
- Exact model
- claude-sonnet-5
- Reasoning
- high
- Agent harness
- none
- Tool access
- none for raw throughput measurement
- Prompting
- Artificial Analysis standard latency suite
Artificial Analysis provider page for Claude Sonnet 5 (high effort) on Anthropic API — output speed ~71 tok/s, TTFT elevated under high reasoning.
Caveats
Known limitations
- High-effort mode is much slower (TTFT) than non-reasoning Sonnet configs.
- AA measurements vary by day and load; treat as order-of-magnitude tempo signal.
- normalisedScore is a curator seed for 0–100 tempo mapping, not AA's own scale.
Weighting
Evidence weight factors
The engine weights each evidence row by these factors (0–1) when fusing scores — higher is more trusted.
relevance0.95
quality0.85
recency0.95
health0.90
comparability0.85
reliability0.75
Benchmark
Artificial Analysis API latency / throughput
Third-party measurements of output tokens/sec and time-to-first-token across API providers for a given model configuration.
Benchmark source →