ApiDelta — May 11, 2026

Takeaway — Anthropic launches Claude API on AWS with native IAM auth; Mistral SDK v2.4.5 ships breaking changes in workflow response shapes.

Three items today: Claude Platform goes live on AWS, the 1M-token beta header is dead on Sonnet 4.5 and Sonnet 4 (>200k requests now error), and Mistral Python SDK v2.4.5 breaks workflow event parsing. Two model retirements land June 15.

[INFO] Claude Platform on AWS: full API via native AWS endpoints

The Claude API is now on Anthropic-managed AWS infrastructure with AWS billing and IAM authentication. Full surface: Messages, Files, Message Batches, Managed Agents, Agent Skills, code execution, tool use — via native AWS endpoints. anthropic-sdk-python v0.101.0 ships a new AWS client class today.

Source | SDK

[URGENT] 1M-token context beta retired on Sonnet 4.5 and Sonnet 4

Since April 30, the context-1m-2025-08-07 header has no effect on claude-sonnet-4-5 and claude-sonnet-4-20250514. Requests over 200k tokens now return an error. Migrate to claude-sonnet-4-6 or claude-opus-4-6 — 1M context is GA there, no header needed.

Source

[HEADS-UP] claude-sonnet-4 and claude-opus-4 deprecated; retire June 15

claude-sonnet-4-20250514 and claude-opus-4-20250514 were deprecated April 14, retirement June 15 — 35 days out. Replace with claude-sonnet-4-6 and claude-opus-4-7.

Source

[URGENT] Mistral Python SDK v2.4.5: breaking changes in workflow response shapes

Two breaking changes: workflows.executions.stream() response data.data.union(CustomTaskInProgressResponse).attributes.payload changed; workflows.events.getstreamevents() response data.data structure changed. Code parsing these directly will break on upgrade.

Source

[INFO] Anthropic Managed Agents: multiagent sessions, outcomes, webhooks in public beta

Since May 6: multiagent sessions and Outcomes are public beta under managed-agents-2026-04-01. Webhooks now cover session and vault lifecycle events. Vault credential background refresh for mcp_oauth is also available.

Source

[INFO] Gemini 3.1 Flash Lite now GA on OpenRouter

Low-latency, high-volume multimodal model — text, image, video, audio, PDF inputs. Available at google/gemini-3.1-flash-lite on OpenRouter.

Source

What you need to do this week


Tooling note — Compare LLM prices across 6 sources, one API call. → OracleForge

📬 *ApiDelta is free forever.* If it saves you time: → Tip the bot: @apideltaalertsbot /tip (Telegram Stars, from 35⭐ ≈ $0.50)

Telegram — instant push

@apidelta_alerts_bot — Premium · 50 ⭐/month · pushed within seconds.

@apidelta_news_bot — Free · one digest per day · your language auto-detected.