Devstral-Small-2 launches as a 24B agentic coding model; Replit expands Google Cloud partnership for enterprise vibe coding; DeepSeek-V3 deploys on NVIDIA GB200 systems; LangSmith adds broadcast trace data features.
Main Content
Devstral-Small-2 Model Release: Launches as a 24B parameter agentic coding model featuring built-in tool use and image understanding capabilities, now live on the platform for developers.
Replit & Google Cloud Partnership: Expands collaboration to leverage more infrastructure and integrate Google models, enabling enterprise-grade "vibe coding" features and broader model access.
DeepSeek-V3 Deployment: Deploys on NVIDIA's GB200 NVL72 system to enable massive-scale inference and optimize scaling bottlenecks for Mixture-of-Experts (MoE) models.
LangSmith Update: Introduces a broadcast feature for code-free transmission of trace data and integrates with OpenRouter to support a wider range of SDKs.
GPT-5.2 on Databricks: Now available via Databricks with support for response APIs and Agent Bricks, enhancing enterprise inference performance on benchmarks like OfficeQA.
Rerank 4 Release: Launches as an advanced reranking model offering top-tier search relevance, speed, multilingual support, and domain-specific understanding.
Elastic Cloud Serverless Optimization: Adopts Google Axion processors to boost indexing throughput by up to 40% and reduce query latency, targeting high-demand AI workloads.