Daily LLM API Advisory – June 3, 2026

🚨 Breaking

• Anthropic – Claude Opus 4.8 (released May 28, 2026) now offers a 1 M token context window by default on the Claude API, Amazon Bedrock, and Vertex AI (200 k on Microsoft Foundry), with a 128 k max output token limit. Existing integrations continue to work, but you should verify that request payloads respect the new limits if you plan to use the larger context.
  URL: https://docs.anthropic.com/en/release-notes/api#may-28-2026
• Weaviate – v1.36.17 & v1.37.7 (released June 2, 2026) list no breaking changes. No code changes are required.
  URL: https://github.com/weaviate/weaviate/releases/tag/v1.36.17

🗑️ Dépréciations

• Anthropic – Claude Sonnet 4 (claude-sonnet-4-20250514) and Claude Opus 4 (claude-opus-4-20250514) are deprecated. Retirement on the Claude API is scheduled for June 15, 2026. Migrate to Claude Sonnet 4.6 and Claude Opus 4.8 respectively.
  URL: https://docs.anthropic.com/en/release-notes/api#april-14-2026

💰 Pricing

• No pricing updates were reported in today’s feed.

🆕 Nouveautés

• Anthropic – Claude Opus 4.8 (see above) – new, most capable GA model.
• llama.cpp releases b9471, b9483, b9487, b9488 (June 2026) add profiler fixes, deprecate llama_set_warmup, support for Qwen 3 SSM architectures, and a BoringSSL update.
  URLs: https://github.com/ggml-org/llama.cpp/releases/tag/b9471, https://github.com/ggml-org/llama.cpp/releases/tag/b9483, https://github.com/ggml-org/llama.cpp/releases/tag/b9487, https://github.com/ggml-org/llama.cpp/releases/tag/b9488
• NVIDIA TensorRT‑LLM v1.3.0rc17 – adds MoT World Model support, multi‑node tensor parallelism for MiniMax‑M2, restores Mistral Large 3 text‑only processor.
  URL: https://github.com/NVIDIA/TensorRT-LLM/releases/tag/v1.3.0rc17
• vllm v0.22.0 – hardens DeepSeek V4, introduces NVFP4 fused MoE support, and includes CUDA optimizations.
  URL: https://github.com/vllm-project/vllm/releases/tag/v0.22.0

💡 Conseil du jour

Action today: Update any production code that still references claude-sonnet-4-20250514 or claude-opus-4-20250514 to the new identifiers (claude-sonnet-4.6 and claude-opus-4-8). Doing this before the June 15 retirement prevents API errors and keeps your pipelines stable.