llm-benchmarks
githubdrose.io

Claude Sonnet 4.6 Benchmarks


Measured on anthropic

335 runs
anthropic site
Avg Tokens / Second
22.90
Decode Tokens / Second
Avg Time to First Token (ms)
1660.00
Runs Analysed
335
Last Updated
Sep 7, 2026, 08:57 PM

Throughput distribution

Throughput over time

Claude Sonnet 4.6anthropic20tok/s

Generated throughput, including reasoning tokens — so a thinking model's line does not collapse; the table ranks on visible tokens · one cell per model, showing its best-covered provider · models with at least 4 samples in the window first, fastest of those at the top · shared vertical scale, 0 to 23 tok/s (99th percentile) · dashed rule is the model's own mean

Every lane measuring this model

2 measured · live first · a lane is a provider, a catalogue id and a transport
ProviderLaneAvg Toks/SecMinMaxAvg TTF (ms)n
anthropicvia OpenRouter22.908.4930.101660.00335
anthropicdirect21.908.9332.302090.0037

Frequently Asked Questions

How fast is Claude Sonnet 4.6?
The latest rolling average throughput is 22.90 tokens per second with an average time to first token of 1660.00 ms across 335 recent runs.
How often are these benchmarks updated?
Benchmarks refresh automatically whenever the monitoring cron runs. The most recent run completed on Sep 7, 2026, 08:57 PM.