Multimodal embeddings at scale: AI data lake for media and entertainment workloads (Mar 12, 2026)
AWS published a solution post demonstrating large-scale multimodal embeddings for media & entertainment using Amazon Bedrock Nova Multimodal Embeddings (amazon.nova-2-multimodal-embeddings-v1) and Nova Pro/Lite for tagging. The post includes architecture, cost examples (batch pricing examples), and important operational details such as the async API concurrency limit of 30 concurrent jobs per account.
Gemini API pricing page updated (wording/typo corrections) — last updated 2026-03-11
The Gemini API pricing documentation was edited and the page 'Last updated' timestamp shows 2026-03-11. Minor wording corrections were applied (e.g., 'fps' -> 'frame' in video input pricing rows and a typo 'FNot available' -> 'Not available'). No pricing amounts were changed in the visible tables.
ChatGPT release notes: GPT-5.1 models retired in ChatGPT
OpenAI Help Center release notes state that as of March 11, 2026, GPT-5.1 models (GPT-5.1 Instant, Thinking, and Pro) are no longer available in ChatGPT. Existing conversations are migrated to corresponding current models (GPT-5.3 Instant, GPT-5.4 Thinking, or GPT-5.4 Pro).
Deprecating the DHE cipher suite for TLS connections
On Mar 11, 2026 Vercel announced it will remove support for the legacy DHE-RSA-AES256-GCM-SHA384 cipher suite on June 30, 2026. After that date, TLS 1.2 clients that only support this cipher may fail to connect; TLS 1.3 and modern clients are unaffected.
Mistral and EcoDataCenter announce large AI data‑center investment in Sweden
News reports in mid-March 2026 indicate Mistral AI and EcoDataCenter are partnering on a ~€1.2 billion investment to build an AI-optimized data center in Borlänge, Sweden (targeting 2027 operations). The project emphasizes European compute capacity and may provide dedicated infrastructure for hosting models or services.
EPO deploys Mistral AI OCR model into patent processing (press release)
The European Patent Office published a press release (11 March 2026) announcing that it has deployed an OCR solution developed with Mistral AI to convert non-machine-readable patent documents into structured data. The system was the result of a proof-of-concept and has been integrated into EPO workflows to improve prior-art searchability and analysis.
langchain==1.2.12 (2026-03-11) adds tracing for wrapped models and tool calls
LangChain published langchain==1.2.12 (11 Mar 2026) which introduces tracing for wrapped models and tool calls. The release adds tracing-related instrumentation to wrap model + tool call flows to capture call traces for observability and debugging of agent executions.
Anthropic announced The Anthropic Institute on its news site (Mar 11, 2026). The Institute will publish research and analysis on societal impacts of powerful AI and expands Anthropic's public policy team and related activities.
Fireworks AI on Microsoft Foundry: per‑token pricing and TPM/PTU quotas
Microsoft announced Fireworks AI is available on Microsoft Foundry with explicit per‑token (serverless) pricing for several models and documented tokens‑per‑minute (TPM) quotas per region/model. The announcement also details provisioned throughput (PTU) minima, scale increments, and input TPM per PTU values for production deployments, plus a 30‑day model retirement notice policy for serverless models.
LangGraph 1.1.0 introduces opt-in type-safe streaming/invoke (version="v2") and migration guide
LangGraph released langgraph==1.1.0 (10 Mar 2026) adding an opt-in version="v2" streaming/invoke format that returns typed StreamPart and GraphOutput objects. The release documents how v2 yields typed dicts/objects, automatic coercion to Pydantic/dataclass schemas, and provides an explicit Migration Guide for incremental adoption. The release notes also clarify backward compatibility: v1 remains the default.
OpenAI Agents SDK docs and blog updates (Mar 2026): skills blog, Agents SDK docs refresh, tool-search & WebSocket guidance
OpenAI updated the Agents SDK developer documentation and published a blog post about using skills to accelerate OSS maintenance in early–mid March 2026. The docs now include expanded guidance for tool search, WebSocket/Responses API usage, skills, and other Agents SDK configuration and runtime topics. The docs site navigation and guides were restructured / refreshed to surface these new topics.
Gemini 3 Pro Preview deprecated and shut down; migrate to Gemini 3.1 Pro Preview
The Gemini 3 Developer Guide states that the Gemini 3 Pro Preview was deprecated and shut down on March 9, 2026. Google advises migrating usage to Gemini 3.1 Pro Preview (and lists alternative 3-series models such as Gemini 3 Flash and Nano Banana variants). The page also notes an update to the Terms of Service on March 23, 2026.
Chat SDK adds table rendering and streaming markdown
On Mar 6, 2026 Vercel's Chat SDK added a Table() card component and updated streaming markdown so adapters convert markdown to each platform's native format during streaming. Streaming now shows formatted content earlier (Slack native streaming) and other adapters run markdown->native conversion during incremental edits rather than receiving raw markdown until completion.
v0 API now supports custom MCP servers (v0 routing change)
On Mar 6, 2026 the v0 API added support for connecting to custom MCP servers, allowing teams to configure and reference custom MCP server IDs in v0 chat sessions.
The openai-agents-python repository published a release series in March 2026 (v0.10.3 on 2026-03-02, v0.10.4 on 2026-03-03, and v0.10.5 on 2026-03-05). Notable changes include adding explicit MultiProvider prefix modes, returning McpError as a structured error result instead of crashing agent runs, various fixes isolating parallel function tool failures, and documentation/compatibility fixes (including a fix for deprecated capwords import for Python 3.11+). Several docs clarifications for realtime transport and advanced model settings were also added.
The openai-agents-js repository published v0.5.3 (2026-03-02) and v0.5.4 (2026-03-05). Key items include exposing agent tool invocation metadata on run results, adding custom rejection messages for approvals, documentation updates (including improved TypeScript docs entry points), and multiple tracing/realtime/docs fixes and dependency bumps.
On Mar 5, 2026 Vercel added openai/gpt-5.4 and openai/gpt-5.4-pro to AI Gateway. GPT-5.4 brings improved multi-step reasoning, faster performance, and better token efficiency compared to prior models.
On Mar 3, 2026 Vercel added google/gemini-3.1-flash-lite-preview to AI Gateway. The model targets high-volume, latency- and cost-sensitive tasks and supports configurable thinking levels (minimal, low, medium, high) and multimodal modalities.
Vercel AI SDK v6.0.108 Released with Ecosystem Updates
Vercel released AI SDK v6.0.108 along with coordinated updates to the full ecosystem including @ai-sdk/react v3.0.110, @ai-sdk/vue v3.0.108, @ai-sdk/svelte v4.0.108, and @ai-sdk/langchain v2.0.114. This batch release ensures cross-framework compatibility.
LangChain released langchain-core v1.2.17, continuing the rapid iteration cycle on the core framework. This is the latest in a series of releases during late February and early March 2026, following v1.2.14 through v1.2.16.
LangChain released langchain-huggingface v1.2.1, updating the HuggingFace provider integration. This ensures compatibility with the latest HuggingFace models and inference endpoints.
Vercel AI SDK received two rapid-fire releases (v6.0.106 and v6.0.107) on March 2, 2026. These releases include incremental bug fixes and improvements to the AI streaming and generation toolkit.
Weaviate released v1.36.2 with fixes for vector index zero-length vector handling and a quantization cache issue. These bug fixes improve stability for production deployments.
Weaviate patch releases (v1.36.2, v1.35.12, v1.34.17) on 2026-02-27 — bug fixes, no breaking changes
Weaviate published multiple patch releases on 2026-02-27 (notably v1.36.2, v1.35.12, v1.34.17) addressing zero-length vector handling, quantization-with-cache issues, and related fixes. The release notes list fixes and performance/backport changes and explicitly show ‘‘none’’ under Breaking Changes. No pricing, licensing, or policy updates were found in the same timeframe.
Anthropic Issues Statement on Secretary of War Pete Hegseth Comments
Anthropic published an official statement responding to comments made by Secretary of War Pete Hegseth regarding AI military applications. The statement includes Anthropic's position on responsible AI use and guidance for its customers.
Dario Amodei Statement on Department of War Discussions
Anthropic CEO Dario Amodei released a statement regarding the company's discussions with the U.S. Department of War. The statement addresses Anthropic's policy stance on government AI partnerships and defense-related use cases.
Weaviate released v1.36.1 with Block Max-Wand performance improvements for faster keyword and hybrid search. The release also includes a fix for replication retry logic.
LangChain released langchain-core v1.2.16 on February 25, 2026, continuing the active development of the core framework library. This mid-week release addresses improvements to the chain and runnable abstractions.
LangChain released langchain-anthropic v1.3.4, updating the Anthropic provider integration. This release ensures compatibility with the latest Anthropic API features and Claude model capabilities.