A quick test to see if Claude Code runs faster using the Anthropic API or the AWS Bedrock API (same model, folder, environment) revealed something shocking. Anthropic API knows double the number of pirate jokes compared to AWS Bedrock. If this is a metric you care about then the results speak for themselves.
Speed Results
Same prompt, same model, 6 runs each:
| Provider | Avg | Min | Max |
|---|---|---|---|
| Anthropic | 5.90s | 5.66s | 6.12s |
| Bedrock | 7.22s | 5.91s | 9.01s |
Anthropic’s API is ~1.3 seconds faster. ~18% for short requests.
Bedrock also had more variance (3.1s spread vs 0.5s for Anthropic), with one outlier run hitting 9 seconds.
Raw data
AWS Bedrock
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why couldn't the pirate play cards? Because he was standing on the deck!
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.26s user 0.44s system 23% cpu 7.341 total
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why couldn't the pirate play cards? Because he was standing on the deck!
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.21s user 0.39s system 17% cpu 9.007 total
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why couldn't the pirate play cards? Because he was standing on the deck!
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.10s user 0.38s system 21% cpu 6.852 total
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why couldn't the pirate play cards? Because he was standing on the deck!
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.27s user 0.41s system 23% cpu 6.988 total
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why couldn't the pirate play cards? Because he was standing on the deck!
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.22s user 0.41s system 27% cpu 5.912 total
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why couldn't the pirate play cards? Because he was standing on the deck!
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.17s user 0.41s system 21% cpu 7.216 total
Anthropic
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why did the pirate go to school? Because he wanted to improve his "arrrr-ticulation."
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.21s user 0.42s system 27% cpu 5.911 total
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why couldn't the pirate play cards? Because he was standing on the deck.
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.27s user 0.41s system 27% cpu 6.017 total
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why did the pirate go to school? Because he wanted to improve his "arrrr-ticulation."
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.26s user 0.44s system 28% cpu 5.877 total
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why did the pirate go to school? Because he wanted to improve his "arrrticulation."
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.25s user 0.41s system 27% cpu 6.121 total
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why couldn't the pirate play cards? Because he was standing on the deck.
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.17s user 0.40s system 27% cpu 5.658 total
$ time claude --print "Tell me a short pirate joke (one or two sentences max)"
Why did the pirate go to school? Because he wanted to improve his "arrrr"-ticulation!
claude --print "Tell me a short pirate joke (one or two sentences max)" 1.32s user 0.43s system 30% cpu 5.804 total
Takeaway
Pretty surprised that Anthropic is faster. Anthropic API goes down sometimes, Bedrock is still a great fallback.