Add Tempr's AWS Bedrock models

[chrissalomon]

Third batch for Tempr's AI Gateway (api.temprhq.io/v1), after #7448 and #7452: the 165 AWS Bedrock models it serves. Tempr relays Bedrock on the caller's own AWS credentials, with no markup on tokens.

Each entry mirrors providers/amazon-bedrock's own base_model attribution, per-model deltas and pricing, since that's the same model served on the same account. Region-prefixed ids (us., eu., global.) are Bedrock's own, and Tempr addresses them as aws-bedrock/<bedrock id>.

Reasoning controls are what Tempr's own GET /v1/models reports for each model, taken from the service's code rather than copied from the lab or the host — the same model family genuinely differs per host. On Bedrock specifically, Tempr sends reasoning_effort on the OpenAI-compatible surface and Converse's own fields on /v1/messages, and Bedrock accepts only each model's declared levels. Each file's leading comment records the exact fields a caller sets; models that reason with no caller control are reasoning_options = [], which on this host means Tempr sends nothing and the x-tempr-reasoning-effort header reports default. Docs: https://temprhq.io/docs/gateway-chat-completions#reasoning

The remaining inference hosts Tempr serves (DeepInfra, Together AI, Fireworks AI, Hugging Face) follow in a separate PR, since those plus this one are past the 300-file limit for one diff.

🤖 Generated with Claude Code

[github-actions]

Action items

• [high] [violation] providers/tempr/models/aws-bedrock/openai.gpt-5.4.toml:19 - Check: Provider files that shrink Bedrock GPT context must drop inherited limit.input (deep-merge + Bedrock peer base_model_omit). Why: Lab openai/gpt-5.4 and openai/gpt-5.5 carry limit.input = 922_000 with context = 1_050_000. These Tempr entries only override limit.context = 272_000, so merge keeps input = 922_000 (> context). providers/amazon-bedrock omits limit.input on the same IDs. Action: Add base_model_omit = ["limit.input"] (and any other Bedrock omits that apply) on openai.gpt-5.4.toml and openai.gpt-5.5.toml.

[github-actions]

No actionable findings.

[github-actions]

No actionable findings.

[rekram1-node]

Too big of a pr.

[chrissalomon]

Thanks — understood. I'll close this and the other Tempr PRs (#7448, #7452, #7587) and resubmit as small per-lab PRs, listing only models that are currently served. The first is #7638, which adds the provider plus Anthropic; the rest will follow one lab at a time once that lands.