Liquid AI has published LFM2.5-24B-A2B, a mixture-of-experts model with 24B total parameters but only 2B active per token, designed to run reasoning agents on local hardware. The release expands the context window to 131K tokens, doubles the vocabulary size, and was trained on roughly double the pretraining data of its predecessor. On agentic and tool-use benchmarks the model performs on par with or better than models that have about 50% more parameters, while delivering fast inference speeds on both consumer and data-center hardware.
Orca Security built Discovery AI, an Amazon Bedrock-based agent that turns plain-English security questions into structured queries against its Sonar system, which tracks over 900 cloud resource types. Rather than relying on one large model with the full schema stuffed into every prompt, the production version splits the work across nine specialized stages and multiple foundation models. The result: dramatically lower cost, faster responses, and better accuracy on complex queries, now running live since March 2026.
AWS detailed three kernel-level optimizations for Mixture-of-Experts (MoE) model serving on Hopper GPUs, implemented in collaboration with the vLLM community and shipped in Amazon SageMaker AI's LMI container. The changes target the inefficiencies of fused MoE computation during low-batch decode, which previously left Tensor Cores underutilized. Combined, the optimizations deliver up to 10% higher throughput and up to 9% faster time-to-first-token (TTFT).
NVIDIA has released the full Run:ai platform as open source, adding it as a foundation of the NVIDIA DSX AI Factory Platform for AI workload and GPU orchestration on Kubernetes. The release includes four modular components — KAI Scheduler, GPU Fractioning Operator, KAI Resource Management, and Workload Map — that can be adopted together or independently. Eight ecosystem partners, including Crusoe, Mirantis, Nebius, OpenNebula Systems, Rafay, Saturn Cloud, Spectro Cloud, and vCluster, are integrating the platform into their own offerings.
Apple ML researchers published a NeurIPS paper introducing SELFCOMPACT, a scaffold that lets language model agents decide for themselves when and how to compact their own accumulated context, rather than relying on fixed-interval summarization triggered by a token count. It combines a compaction tool the model can call with a lightweight rubric telling it when to fire (a sub-task is resolved) and when to hold off (mid-derivation or stuck). Tested on six benchmarks and seven models, it matched or beat fixed-interval summarization while using far fewer tokens.
Amazon Web Services researchers published a method called 'priming' that converts pretrained Transformer models into hybrid architectures mixing attention and state space model (SSM) layers, reusing existing weights instead of retraining from scratch. The approach cuts pretraining costs by over 99% and lets a 32-billion-parameter hybrid model match its source Transformer's quality while decoding up to twice as fast. Amazon tested three SSM types and released an open-source model zoo, training code, and serving tools under Apache 2.0.
Mistral AI has released Voxtral Mini 4B Realtime Arabic, a streaming speech-to-text model fine-tuned for Arabic dialects, Modern Standard Arabic, and code-switching between languages. The 4.4-billion-parameter model processes 16 kHz audio in real time with a configurable transcription delay, achieving an 8.82% average Character Error Rate across seven Arabic benchmarks at a 480ms delay — close to the offline Voxtral Transcribe Arabic model's 7.91% CER. It was co-developed with Morocco's digital transition ministry as part of a sovereign AI partnership and is released under Apache 2.0.
Anthropic has introduced a beta option on its Claude Platform that allows a Managed Agents agent to construct its own workflow for large, multi-part jobs, such as going through hundreds of documents. Instead of handling everything in one pass, the agent splits the task into phases, delegates pieces across multiple sub-agents, and merges their output into a single result. The job is handed off to the server, which executes it asynchronously and reports progress through a stream of events.
AWS published a joint engineering write-up with Postman describing how Agent Mode, Postman's AI-native interface for API testing and documentation, was built to run in production at scale. Rather than model quality, the team found that tool sprawl and missing context were the biggest obstacles to reliable agent behavior. The post details architectural patterns including dynamic tool scoping, schema-based query access, dedicated context handlers, and how Amazon Bedrock provides model flexibility, cross-Region inference, data residency controls, and prompt caching.
Google released ML Drift, an open-source, cross-platform GPU compute engine for on-device ML inference, replacing the legacy TFLite GPU delegate. It unifies shader handling across OpenGL ES, OpenCL, Metal and WebGPU, adds 5D tensor support, and introduces stage-aware optimizations for LLM prefill/decode. It is already running in production in Chrome, YouTube Shorts, Google Photos, Meet and partner apps from Adobe and Snap.
Google introduced AQuA (Ambient Quality Agent), a tool that runs continuously beside production AI agents inside a Google Cloud project to find failures that don't show up in logs or health checks, such as skipped validation steps or dropped context between sub-agents. It samples production session traces, grades them against a checklist, clusters recurring failure patterns, verifies clusters against full transcripts, and tracks them over time as new, recurring, or resolved. On request, it can also trace a failure back to the exact line of code or prompt responsible, using an immutable source snapshot captured at deploy time.
Google Cloud announced the Gemini agent, a unified agent that combines chat, knowledge work, content creation, and coding into one interface accessible across devices and channels. It integrates directly inside Gmail, Docs, Sheets, Slides, Chat, and Calendar, connects to enterprise tools and databases via MCP and a tools registry, and orchestrates across multiple models — including Google's Gemini family and Anthropic's Claude — to balance quality and cost. Google also detailed security/governance features, cost controls like Smart Routing and spend caps, and shared customer results from companies like BNP Paribas, Bradesco, and Orange Spain.
Google has released the Developer Knowledge API, a structured, machine-readable source for Google Cloud, Firebase and Android documentation aimed at AI agents and developer tooling. Instead of scraping web pages or relying on an LLM's training cutoff, agents and CLIs can now query fresh, Markdown-formatted docs through several interfaces: a gcloud CLI surface, an installable agent skill, client libraries, and a browser-based API explorer. The underlying index is refreshed frequently so retrieved answers stay close to the latest published documentation.
AWS describes a new two-stage access control architecture for Retrieval Augmented Generation (RAG) in Amazon Quick and Amazon Bedrock Knowledge Bases. Instead of relying only on periodically synced access control lists (ACLs), the system now verifies permissions in real time against the authoritative data source (e.g. Google Drive, SharePoint, Confluence) at query time, before passing content to the LLM. AWS says this closes security gaps caused by stale or incorrectly mapped permissions between sync cycles. Mondelēz International, with over 35,000 employees, is cited as a customer using this approach.
Anthropic has launched Claude Haiku 5.5 (claude-haiku-5-5), positioned as its most capable Haiku model for high-volume, latency-sensitive work. It supports a 1 million token context window, up to 128k max output tokens, and adaptive thinking controlled via a new effort parameter. The model is available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry. Anthropic warns that code written for the previous Haiku 4.5 model may break on the new version.
Cloudflare built an AI agent system for its Managed Defense security service to handle the flood of alerts that overwhelm human analysts. Instead of one general-purpose agent analyzing everything, the system separates deterministic data collection from model-based reasoning and runs four specialist agents in parallel before a synthesis agent combines findings. The company says this reduces hallucinations and makes investigations auditable and reproducible.
The Technology Innovation Institute (TII) has released Falcon-ASR, a 1.6 billion parameter automatic speech recognition model built primarily for Arabic, with particular focus on the Emirati dialect. It also transcribes English, French, Spanish and Portuguese using the same model weights. TII reports it beats the best published leaderboard result on a standard Arabic benchmark and leads on an internal Emirati evaluation, while also including word-level timestamps.
Deepgram has expanded its medical-specialized speech model, Nova-3 Medical, from English-only to 10 languages including Spanish, German, Hindi, Japanese and Russian, with automatic code switching in both batch and streaming modes. On a 115-hour curated medical audio test set, the multilingual medical model cut word error rate by 25-30% and entity error rate (drugs, doses, conditions) by 37-40% compared to Deepgram's general-purpose multilingual model. It also beat competing models from ElevenLabs and Microsoft on average word error rate across all 10 languages.
AssemblyAI introduced the Sync API, a new transport for transcribing short audio clips: one HTTP POST request returns a finished Universal-3.5 Pro transcript in the same response, at roughly 134ms median latency. It fills the gap between the Async API (submit-and-poll, 5-6 seconds of added latency) and the Realtime API (WebSocket, built for ongoing sessions), targeting workloads like dictation, voice-agent turn transcription, IVR, and push-to-talk. Clips from 80ms to 2 minutes and up to 40MB are supported, with word error rate of 1.59% on short-form audio. Pricing is $0.45/hr, the same rate as Universal-3.5 Pro Realtime.
AssemblyAI launched a hosted deployment of the open-source Qwen3.5 4B model on its LLM Gateway, optimized specifically for rewriting raw speech-to-text output into clean, formatted text. On voice rewrite tasks like dictation cleanup and transcript formatting, the model averaged 612ms response time—1.9x faster than GPT-4.1—at 94% lower cost per hour of audio processed. The model is priced at $0.10/$0.50 per million tokens (prompt/completion) and supports only max_tokens, temperature, and stream parameters, with no tool calling.
AssemblyAI released Universal-3.5 Pro, a new flagship async speech-to-text model priced at $0.21/hr. It natively transcribes code-switched conversations across 18 languages without separate configuration, introduces jointly modeled speaker diarization built directly into the transcript (rather than stitched from a separate system), and supports contextual prompting to prime the model with domain knowledge. Benchmarks show lower word error rate on code-switched audio and higher cpWER accuracy on diarization compared to several competitor models and AssemblyAI's own prior Universal-3 Pro.
Google has launched EmbeddingGemma 2, a Gemma 4-based embedding model released under Apache 2.0 that places text, code, images, video and audio in one shared 768-dimensional vector space. It's modular, so developers load only the encoders they need: 270M parameters for text and code alone, scaling up to 740M for every modality. Google reports a 14% jump over EmbeddingGemma 1 on the MTEB Code benchmark while keeping the earlier model's multilingual text accuracy.
Z.ai's GLM 5.3, a 753-billion-parameter mixture-of-experts model optimized for coding and long-horizon agentic tasks, is now available as a fully managed model on Amazon Bedrock. It builds on GLM 5 with improved coding benchmark results, standout cybersecurity capabilities (CyberGym score of 84.5), and deeper Bedrock integration including OpenAI-compatible APIs, prompt caching, cross-Region inference and service tiers. AWS also demonstrates using GLM 5.3 with the open-source Strix agent for authorized penetration testing of your own applications.
GitHub has published ReviewBench, a research-preview benchmark designed to compare AI code review agents on a common, auditable standard. The dataset draws on 219 real public pull requests across 19 languages and 187 repositories, chosen to mirror the language and size distribution found across 103.9 million GitHub pull requests. Ground truth comes from a blended pool of human reviewers, static analysis tools, and multiple frontier LLMs, cross-checked by senior engineers who agreed with the benchmark's labels 96.6% of the time. GitHub says it has already used ReviewBench internally to track its own Copilot code review (CCR) product, and that offline benchmark movement reliably predicted the direction of later production A/B tests.