[brmur]

This pull request updates the Converse and ConverseStream Anthropic Claude samples across six languages (plus the Go unit test) to a current, working model. It is the multi-language follow-up to #8204, which fixed the Python samples behind the same documentation feedback.

Why

These samples pinned anthropic.claude-3-haiku-20240307-v1:0, which has reached end of life — running them as published fails with a 404 (ResourceNotFoundException, "reached the end of its life", re-confirmed in us-east-1).

What changed

• Repoint to the active Claude Haiku 5.5 inference profile us.anthropic.claude-haiku-5-5 (status ACTIVE; the us. prefix matches the us-east-1 Region these samples use).
• Remove the temperature and topP inference parameters from the Java, .NET, JavaScript, and Swift samples. Claude Haiku 5.5 deprecates both temperature and top_p entirely (ValidationException: `temperature` is deprecated for this model / `top_p` is deprecated for this model), so a model-ID swap alone would leave those samples broken. The Rust and Go samples set no inference config and are unaffected.
• Update the Go unit-test constant CONVERSE_CLAUDE_MODEL_ID in lockstep with converse.go so the stubbed test still matches.

Files (15)

• Java v2: Converse.java, ConverseAsync.java, ConverseStream.java
• .NET v3: Converse.cs, ConverseStream.cs
• .NET v4: Converse.cs, ConverseStream.cs
• JavaScript v3: converse.js, converseStream.js
• Rust v1: converse.rs, converse-stream.rs
• Swift: Converse/main.swift, ConverseStream/main.swift
• Go v2: actions/converse.go + actions/converse_test.go

Verification

• Each inference-config shape used by the samples was exercised live against us.anthropic.claude-haiku-5-5 and returns 200:
  • maxTokens only (Java/.NET/JavaScript, after removing temperature)
  • maxTokens + stopSequences (Swift, after removing temperature)
  • no inference config (Rust/Go)
• gofmt and cargo fmt --check are clean on the modified Go and Rust files.
• Snippet markers are unchanged; Go/Rust line counts are unchanged (1-for-1 line edits), so README line-number anchors remain accurate.

Not included

Two tool-use demos (rustv1/.../tool-use.rs and python/.../tool_use_demo/tool_use_demo.py) also reference retired IDs but exercise a different API surface (and the Python one carries additional retired Claude 3 Opus/Sonnet IDs in a SupportedModels enum). Those are intentionally deferred to a separate change so tool use can be verified independently.

Related: internal ticket V2245921888.

---

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[github-actions]

🤖 AI Code Example Review

This PR passes review. It makes a focused, well-reasoned model-ID update across 15 files to replace an end-of-life Claude 3 Haiku model with the active Claude Haiku 4.5 inference profile, and correctly removes the topP inference parameter that the new model rejects when combined with temperature. All changes are consistent, commented, and the Go unit test constant is kept in sync.

Detailed Review

1. [nit] dotnetv3/Bedrock-runtime/Models/AnthropicClaude/Converse/Converse.cs and its dotnetv4 counterpart both contain a duplicated namespace AnthropicClaude { public partial class Converse { } } declaration at the end of the file. This is a pre-existing issue not introduced by this PR, but worth flagging since it will generate a compiler warning about duplicate partial class declarations.

2. [nit] The Go converse.go file declares modelId as a local variable (modelId := "us.anthropic...") rather than a package-level or file-level constant. This is inconsistent with the Rust files, which use const MODEL_ID. Since the model ID is not dynamically determined, a const would better signal its fixed nature and be consistent across languages. That said, this is a pre-existing pattern and not introduced by this PR.

3. [nit] The Swift Converse/main.swift and ConverseStream/main.swift files retain stopSequences: ["END"] in their InferenceConfiguration. This is a pre-existing parameter not changed by this PR, and it's worth confirming it does not cause a ValidationException with Claude Haiku 4.5 in the same way topP did. The PR description mentions the Swift shape was tested live and returns 200, so this appears verified.

4. [policy] No new automated tests were added for the non-Go languages (.NET, Java, JavaScript, Rust, Swift). The Go unit test constant was correctly updated to match the new model ID. For the other languages, the PR relies on live manual verification rather than CI-enforced tests. This is acceptable for a model-ID-only change, but worth noting as a policy observation.

5. [nit] The README.md line-number anchors in gov2/bedrock-runtime/README.md and rustv1/examples/bedrock-runtime/README.md were updated to reflect the new line positions after adding comment lines. This is good housekeeping. However, README anchor line numbers are fragile and will drift with future edits — consider whether the project has a better mechanism (e.g., named anchors) for this.

6. [nit] The comment style for the model ID explanation is slightly inconsistent across languages. Java, .NET, and JavaScript use a two-line comment block; Go and Rust use a single-line comment; Swift uses a two-line comment. This is a minor cosmetic inconsistency across the codebase and is not introduced by this PR.

7. [policy] The PR description notes that two tool-use demos referencing retired model IDs are intentionally deferred. This is a reasonable scoping decision, but the deferred work should be tracked to avoid those samples remaining broken indefinitely.

8. [nit] In converse.go, the comment reads // the "us." cross-Region profile ID. without explicitly mentioning that it matches a specific region. The other language files say ...that matches the Region set above or ...that matches CLAUDE_REGION below, which is more informative for readers. The Go version is slightly less explicit.

9. [nit] The CONVERSE_CLAUDE_MODEL_ID constant in converse_test.go uses UPPER_SNAKE_CASE, which is idiomatic for Go exported constants but the name is arguably redundant with the CONVERSE_ prefix given the file context. This is pre-existing style and not introduced by this PR.

10. [policy] The PR has been verified manually against us-east-1, but the inference profile ID us.anthropic.claude-haiku-4-5-20251001-v1:0 is region-specific (the us. prefix). Samples that hardcode us-east-1 as the region are consistent with this profile, but readers deploying in other regions (e.g., eu-west-1) would need a different profile prefix. A brief note in the comments or README about this constraint would improve discoverability, though this is a minor documentation nit.

---

_{Reviewed at commit 04bf88c4d5fa59a9128df608de1158c0e9a7360c. This review was generated automatically using Amazon Bedrock. It compares your changes against existing examples and coding guidelines. Findings are labeled by severity (blocking / nit / policy) and reflect only the commit above. Please use your judgment — this is advisory, not authoritative.}

[brmur]

Depends on #8207 (merge that first).

The Lint Rust check on this PR fails for a pre-existing, repo-wide reason unrelated to these changes: CI pins rustc 1.88.0 but the current AWS SDK for Rust crates require 1.94.1. #8207 bumps the toolchain and fixes the resulting lint/dependency failures. Once #8207 merges to main, rebasing this PR will clear its Lint Rust check.

[DennisTraub]

One change requested across all examples: Use us.anthropic.claude-haiku-5-5

[DennisTraub]

LGTM