Large Language Model News

Redis launched LangCache, a fully managed semantic caching service now in public preview on Redis Cloud, which Redis reports can cut LLM API costs by up to 90% and return cache-hit responses up to 15x faster than re-quer…

A former B2B SaaS web developer who switched to inference engineering about a month ago published a skills map arguing the field requires only three service categories: routing and scheduling via Nvidia Dynamo or llm-d, …

Anthropic's September report documents that DeepSeek, Xiaomi, and Moonshot fed conversations between their own models and users into Claude and used Claude's responses as training data to distill its capabilities, with s…

Anthropic said Thursday it disrupted several attempts this year by researchers to use its Claude AI models for biological research that could aid biological weapons development, banning the associated accounts and sharin…

A developer tested whether a frontier LLM could reverse-engineer secrets from a hardened, decompiled application build. The build XOR-encodes configuration values using working keys derived at runtime from a single 32-by…

A user identified as Mateo-Grgic posted a request to add DeepSeek V4.1 Flash, describing it as a new model that is better and cheaper than the existing DeepSeek V4.1 Flash. The post contains no further details, figures, …

A developer released VernLLM, an open-source MIT-licensed LLM call framework for TypeScript, and stress-tested it with over 1,000 LLM calls. The framework offers adaptive AIMD rate-limiting, advanced circuit-breakers, mu…

Anthropic said it disrupted several cases this year in which scientists used its Claude AI model for biological research that could potentially aid bioweapons development, according to a report the company published Thur…

GraphQL, introduced in 2015 as a typed, introspectable and composable API contract, is being positioned as the API protocol for AI-native systems, according to the article. The piece argues GraphQL fits the Model Context…

Amazon SageMaker Inference launched prefix-aware routing, a routing strategy that directs requests sharing the same prompt prefix to the same instance so cached key-value pairs are reused. Benchmarks on Llama 3.1 70B Ins…

Recursive self-improvement (RSI) is a hypothesized process in which artificial general intelligence (AGI) systems rewrite their own computer code, potentially triggering an intelligence explosion that could result in sup…

Developers can stop large language models from hallucinating nonexistent API calls by injecting local documentation or current SDK source into the context window rather than relying on the model's training data, accordin…

Anthropic's latest threat intelligence report disclosed that state-linked actors in Mali, China and Iran used its Claude models — specifically Claude Haiku, Sonnet and Opus — to automate surveillance work, including mobi…

Osprey, a target-agnostic speculative decoding method detailed in arXiv:2609.09338v1, raises mean acceptance length by 16.1% on Qwen3-8B, 21.2% on Llama-3.3-70B-Instruct, and 22.7% on MiniMax-M2.5 (229B), with a 17.5% ga…

Amazon launched model caching for Amazon SageMaker Inference on HyperPod, a feature that pre-loads model weights and container images onto cluster nodes so inference pods can read from local NVMe storage at approximately…

An independent researcher is seeking an arXiv cs.AI endorsement to submit a first paper titled "HindsightTag: A Synaptic Tagging-and-Capture Framework for Retroactive Memory Consolidation in LLM Agents." The paper formal…

Bedrock Robotics posted an internship opening for a 2027 Onboard Infrastructure Engineer, ML Inference role based in San Francisco, CA, on a hybrid schedule, with pay not published. The intern will integrate open-source …

Prompts.chat emerged as an open-source repository that turned natural language prompts into a version-controlled, multi-platform library after ChatGPT's late-2022 debut, evolving from a raw Markdown collection into an ex…

Developer Sergi Corruchaga released D-Engine v0.2.2, an open-source coding harness that replaces the agentic loop with a deterministic two-call pipeline, cutting token usage by up to 42× versus DeepSeek's official agent …

DeepSeek's V4.1 Flash scored 81.2 out of 100 on OpenDesign Arena's real-world design benchmark, reaching 98% of OpenAI GPT-6 Astra's top score of 82.7 while costing $0.023 per finished design versus Astra's $1.61, accord…