You'll noticeTier: how much it should matter to you
3Useful: my rating, 1 to 5
1Signal: worth watching, 1 to 5
LLM GatewayArea: what it touches
ImprovementsKind: in v2.1.295,
ImprovementsSection of the release
What
A gateway is a proxy between Claude Code and the model provider. Bedrock is Amazon's model service. count_tokens asks how many tokens (pieces of text the model counts) a request uses.
The Bedrock upstream now answers count_tokens with AWS CountTokens, with a 10-second timeout, instead of failing at once.
If that fails it retries only every 10 minutes, and meanwhile falls back to the next upstream or to a billed one-token request.
The built-in gateway guide now says to answer count_tokens with Bedrock's CountTokens and return {"input_tokens": N}, using 501 only if that fails or the model is an application inference profile. Before, it said Bedrock had no count-tokens API.
Why
Token counts on Bedrock gateways can now come from Amazon directly. The AWS identity the gateway uses needs permission for bedrock:CountTokens, or it falls back as described.
