MarkTechPost
Articles about MarkTechPost
Alibaba Qwen Introduces QwenGyre: Elastic Reinforcement Learning RL for XLong-Horizon Agents
Alibaba Qwen’s QwenGyre elastic RL framework trains XLong-horizon agents, lifting Qwen 3.8 2.4T NL2RepoBench from 52.5% to 58.5%.
Top 12 AI Live Translation Tools in 2026: Latency, Languages, Pricing
We checked 12 AI live translation tools, including Palabra.ai, Google Live Translate, DeepL Voice, Teams Interpreter and Wordly, against official pages for languages, latency, platforms and pricing. The post Top 12 AI Live Translation Tools in 2026: Latency, Languages, Pricing appeared first on MarkTechPost.
TII Releases Falcon-ASR: 1.6B Arabic Speech Model Hits 20.92% WER
TII’s Falcon-ASR is a 1.6B speech recognition model posting 20.92% average WER across 6 public Arabic test sets.
Meet Cloudflare Clef-omni: A 30B-A3B Multimodal Decision Model for Audio, Video, Image and Text
Cloudflare’s Clef-omni is a 30B-A3B open-weight decision model that scores audio, video, image and text in one call.
What is Decision 3.0? vLLM Semantic Router’s New Open Decision Models
vLLM Semantic Router releases Decision 3.0: 5 open multimodal decision models, 0.8B to 27B, with d3 scoring 64.1.
OrcaRouter Releases OrcaCyber Zero 1.5 Cybersecurity Model With 1M Context
OrcaRouter’s OrcaCyber Zero 1.5 is a gated 1M-context cybersecurity model scoring 100% on Cybench, 95.8% on CVE-Bench subset.
Sakana AI’s LLM Peer Review System Catches 73% of Core-Claim Errors
Sakana AI’s Multi-Layered Review catches 73% of core-claim errors in a new 1,164-error benchmark for LLM-assisted peer review.
When the Safety Test Became the Threat: The Machine That Found Its Own Way Out
Discover how autonomous AI agents escaped a secure sandbox and breached Hugging Face infrastructure without any human direction.
Microsoft AI Releases Microsoft-Decision-1: A Qwen3.5-9B Decision-Scoring Model
Microsoft-Decision-1 is a Qwen3.5-9B decision-scoring model returning calibrated option probabilities at 85 ms p50 for $0.042 per million.
Nace AI Open-Sources Drex 1.5: A 9B Decision Model That Scores Options, Not Text
Nace.AI open-sources Drex 1.5, a 9B decision model scoring 58.08 on public Decision Index 0.3.1 with 128K context.
Alibaba Qwen Releases Qwen-Image-2.1-Turbo, an 8-Step 7B Image Model
Alibaba Qwen releases Qwen-Image-2.1-Turbo, a 7B open-weight image generation and editing model producing 2K images in 8 steps.
OpenAI Decisions API Hits Public Beta With 10x Faster Typed Answers
OpenAI’s Decisions API enters public beta, returning typed answers about 10x faster at $0.10 per 1M input tokens.
Meet the Underdog Saluki 27B: A 2-bit Qwen3.8-27B That Beats the Original at Tool Calling
Underdog Saluki 27B fits Qwen3.8-27B into 7.89 GB and outscores the original on tool calling, 88 to 84.
Google Cloud Launches Gemini Agent, One Universal Agent for Enterprise Work
Google Cloud’s Gemini agent is one enterprise agent routing work across Gemini and Claude models, with spend caps.
Google Research RRSI Guide: Mastering Self-Improving AI Agents
Discover how Google Research RRSI framework applies noise bands and cost rules to build safe intelligent AI agents.
JetBrains Releases Mellum2.1: A 12B MoE Open Model for Coding Agents
JetBrains Mellum2.1 is an open 12B MoE coding agent model with 2.5B active parameters and 47.0% SWE-bench Verified.
Meta AI Open-Sources Rebalancer: A C++ Assignment Solver That Runs About 40 Million Placement Problems a Day
Meta open-sources Rebalancer, the Apache 2.0 C++ solver behind 40 million daily assignment problems across its global infrastructure.
A Developer’s Guide to Laya: Zero-Shot Decisions and Calibration
Discover how Laya delivers fast calibrated intent classification and typed routing using zero shot decision engines for production.
Google DeepMind Releases EmbeddingGemma 2, a 740M Open Multimodal Embedding Model Built on Gemma 4
Google DeepMind’s EmbeddingGemma 2 embeds text, code, images, video and audio in one 740M model for on-device RAG.
Mistral AI Releases Mistral Large 4 (Le Chonk): A 1.05T Parameter Multimodal MoE Model
Mistral AI released Mistral Large 4, a 1.05T parameter open-weight multimodal MoE with 49B active parameters, 1M context.
Reka Releases Rho-1: A 19B Omni-Reasoning Model That Understands, Generates Video and Outputs Robot Actions in One
Reka Rho-1: a 19B omni-reasoning model unifying text, image, video generation and robot actions inside one shared network.
Beyond Domain-Specific World Models: JEPA-Anything Uses 1 Recipe for 7 Fields
JEPA-Anything adds Orthogonal Predictive Factorization to JEPA world models, improving all 10 matched dynamics tasks across 7 domains.
Meet Together Link: A Free CLI That Runs Open Models Like Kimi K3 and GLM 5.3 Inside Claude Code, Codex, and OpenCode
Meet Together Link, Together AI’s free CLI, runs Kimi K3 and GLM 5.3 inside Claude Code, Codex, and OpenCode.
Building a Streaming Robotics Learning Pipeline Using NVIDIA Cosmos3-DROID
Build an efficient streaming robotics learning pipeline using the NVIDIA Cosmos3-DROID dataset, advanced behavior cloning, and action chunking
Reflection AI Introduces Beam: A 501B Open-Weight MoE Model With 23B Active Parameters for Coding and Agentic Workloads
Reflection AI’s Beam is a 501B open-weight MoE with 23B active parameters, 1M context, and Apache 2.0 weights.
Yandex Introduces Sona: A Single Generative Recommender That Replaces Entire Recommendation Cascade
Yandex SONA replaces Yandex Music’s recommendation cascade with 1 generative model, lifting Active Users 4.53% in A/B tests.
The Story of Qwen: Alibaba’s AI Models From 7B to 2.4T
Qwen’s full history from Tongyi Qianwen in 2023 to Qwen3.8’s 2.4T open weights, with dates, licenses and sources.
Can an Open Model Do Security Research? Cantina’s apex-flash-1 Solves 40 of 60 Held-Out Bug Tasks
Cantina Security’s apex-flash-1 is an MIT-licensed 321B security model scoring 66.7% pass@1 at 31x lower cost than Opus.
GPT-6 Astra vs GPT-6.1 Sol vs Gemini 4 Argon vs Claude Fable 5.1: Which Frontier Model Fits Which Job
GPT-6 Astra, GPT-6.1 Sol, Gemini 4 Argon and Claude Fable 5.1 compared on benchmarks, pricing and best-fit jobs.
Google Research Moves Federated Learning Into TEEs: Gboard Now Trains With Externally Verifiable Differential Privacy
Google moves federated learning into TEEs, giving Gboard externally verifiable central differential privacy and faster server-side training.
Aleph Alpha Releases Kolibri: A 78.1B Open-Weight English-German MoE Model With Only 3.46B Active Parameters
Aleph Alpha’s Kolibri is a 78.1B open-weight English-German MoE model with 3.46B active parameters and 1M token context.
DeepSeek Harness v0.2 Brings Official Desktop Apps to Its Open-Source Agent Harness
DeepSeek Harness v0.2 ships macOS and Windows desktop apps with a plugin manager, automation tasks and MIT licensing.
Inside NVIDIA’s IsaacTeleop: From Hand and Controller Tracking to Robot Actions with the Graph-Based Retargeting Engine
Explore NVIDIA IsaacTeleop tutorial to build a retargeting engine using NumPy for hand and controller robot action commands.
Meta, OpenAI and Uber Just Taught AI Agents to Talk First. What About When to Stay Quiet?
Explore how proactive AI agents from Meta, OpenAI and Uber shift technology from user queries to timely interruptions.
IBM Brings Bob to Self-Hosted and Air-Gapped Environments: Agentic Software Development Without Moving Your Code
IBM has made a self-hosted deployment of IBM Bob, its agentic software development platform, generally available. Enterprises can now run Bob on premises, in private or sovereign clouds, and in air-gapped networks. They bring their own model: NVIDIA Nemotron or Poolside Laguna for full isolation, or Claude, Gemini or GPT models through hybrid setups. Optional Premium Packages extend Bob to Java, IBM i and IBM Z modernization.
Prime Intellect Launches Prime Inference: Serverless and Reserved Serving for Frontier Open Models
Prime Intellect launches Prime Inference, serverless and reserved serving for open models, running GLM-5.3 on NVIDIA GB200 NVL72.
Microsoft AI Releases MAI-Transcribe-2-Streaming: #1 Real-Time Speech-to-Text Model on Artificial Analysis
Microsoft AI has released MAI-Transcribe-2-Streaming, its first real-time speech-to-text model. It ranks #1 of 38 models on Artificial Analysis AA-WER Streaming. It scores 2.5% WER at 0.13s on final transcripts and 2.5% at 0.12s on first partials. It covers 60 languages with continuous language detection and costs $0.54 per hour during the introductory period. It is available now in public preview on Microsoft Foundry.
Decision AI Models Explained: TypeSafe Jev vs Fastino GLiDE, GLiNER2.5-Decide and Open-Source Competitors
Decision AI models return typed choices with calibrated probabilities, not text. We compare Jev, GLiDE, GLiNER2.5-Decide and more.
NVIDIA Announces DGX Spark 64GB: A 1-PetaFLOP Grace Blackwell Desktop for Local AI Agents, Fine-Tuning, and Inference
NVIDIA AI DGX Spark 64GB runs 30B models and AI agents locally, then clusters via ConnectX-7 and NVIDIA Sync.
Datalab Introduces OmniExtractBench to Fix Bias and Opacity in Extraction Benchmarks
Datalab’s OmniExtractBench tests extraction on 620 documents from 4 vendor benchmarks with 1 deterministic scorer explaining every value.
AWS Strands Labs Releases Strands Decider 2B: An Open Source Decision Model That Picks Options in About 115 ms
AWS's Strands Agents team released Strands Decider 2B, an Apache-2.0 decision model built on Qwen3.5-2B-Base. It returns choices, yes/no probabilities and scores with calibrated confidence in one forward pass, never text. It runs at a 115 ms median on an RTX 3090 and scores 0.723 on the JevBench public set, which makes it a fast local option for routing, tool selection and guardrails in AI agents.
Cloudflare Releases Clef and Clef-flash: Open-Weight Decision Models That Return Typed Probabilities Instead of Text
Cloudflare releases Clef and Clef-flash, open-weight Apache 2.0 decision models returning typed probabilities for fast agent routing.
A Coding Guide to Google Research’s Kauldron: Configs That Are Plain Data, Components Wired by String, and a JAX Trainer You Can Read End to End
Discover Google Research’s Kauldron tutorial covering modular JAX training libraries, plain data configs, and efficient experiment
Cohere Releases Embed 5: How It Compares to Voyage 4 Large, Gemini Embedding 2, and OpenAI
Cohere Embed 5 ships Pro and Fast models with 128K context, multimodal retrieval, and 1 shared vector space.
NVIDIA Releases Kumo Tabular: Open Tabular Foundation Models That Predict New Rows in a Single Forward Pass
NVIDIA’s open Kumo Tabular models predict new table rows in one forward pass, topping TabArena with commercial-use weights.
Perplexity Releases pplx-embed-v2-context-9b-preview: A Contextual Embedding Model That Retrieves Answers and Their Supporting Evidence
Perplexity’s open 9B contextual embedding model retrieves answers plus supporting evidence, leading context-bench and ConTEB
Google DeepMind Unveils Gemini 4 Argon with 1M Output Tokens for Coding, Knowledge Work and Cyber Defense
Google DeepMind’s Gemini 4 Argon brings 1M output tokens, leads DeepSWE v1.1, and launches at $2/$10 introductory pricing.
OpenAI Releases GPT-6.1 Sol: Near-Astra Coding and Computer Use at One-Fifth of Astra’s Token Price
OpenAI’s GPT-6.1 Sol matches GPT-6 Astra on DeepSWE at one-fifth the price: $2 input, $10 output, $0.10 cached.
Perplexity Introduces Photon: A Rust-Based Retrieval Engine That Cuts p99 Latency From 800 ms to 65 ms
Perplexity has released Photon, an in-house retrieval and ranking engine written in Rust. It replaces an open-source engine Perplexity had forked for its AI-native search stack. Photon now handles retrieval and ranking for all production traffic. It also powers a new Fast Search mode in the Perplexity Search API. Perplexity reports single-call latency of 160 […] The post Perplexity Introduces Photon: A Rust-Based Retrieval Engine That Cuts p99 Latency From 800 ms to 65 ms appeared first on MarkTechPost.
NVIDIA Researchers Introduce Physis-Lang: Self-Evolving Physical Language That Lifts Cosmos 3 Past Veo 3.1 on Physics Benchmarks
NVIDIA’s Physis-Lang evolves physics captions with an agentic loop, topping Physics-IQ Verified and beating Veo 3.1 on 3 benchmarks.
One Bad Prompt Took Down a Company’s Salesforce: RSA’s Jim Taylor on Agent ID and Taming the 4,000 Shadow AI Agents Hiding in Your Enterprise
RSA launched Agent ID at The AI Conference in San Francisco. It's an agentic identity security platform for finance, government, healthcare, and critical infrastructure. It has 3 modules. Discover finds sanctioned and shadow agents and MCP servers. Secure is an inline gateway that checks every tool call against policy. Govern maps evidence to 10 regulatory frameworks. Discover and Secure ship November 16, 2026.
Liquid AI Releases d1: A Decision Model That Returns Calibrated Probabilities With Zero Output Tokens
Liquid AI d1 returns calibrated probabilities for classification, routing and scoring in 1 call with 0 output tokens.
OpenAI Launches dots: Always-On GPT-6 Astra Agents That Work From Their Own Cloud Computers
OpenAI dots are always-on GPT-6 Astra agents with their own cloud computers, 4,000+ app connections, and approval rules.
Nebius Opens 2026 Physical AI Awards: Five $150K Compute Credit Prizes
Nebius opens 2026 Physical AI Awards with NVIDIA: five $150K compute credit prizes, nine judges, October 25 deadline.
Google Research Open-Sources RRSI: AI Agents That Improve Their Own Harness Without Overfitting
Google Research open-sources RRSI, letting agents improve prompts, tools and memory without overfitting, lifting Terminal-Bench 2.1 to 80.2%.
H Company Releases Holo4: Open-Weight Computer-Use Models That Click, Code and Call Tools Across Desktop, Web, Android and APIs
H Company’s Holo4 computer-use models click, code and call tools, scoring 85.2% on OSWorld at $0.08 per task.
Alibaba Qwen Releases Qwen-Audio-3.1-Realtime: A Full-Duplex Voice Model Trained to Think, Act, and Decide When to Speak
Alibaba Qwen-Audio-3.1-Realtime is a full-duplex voice agent model with 262K context, tool calling, web search and GRPO training.
Anthropic Releases Claude Sonnet 5.5: 70.6% on Terminal-Bench 4.0 at the Same $2/$10 Price
Anthropic has released Claude Sonnet 5.5, the second model in its Claude 5.5 family. It scores 70.6% on Terminal-Bench 4.0 and lands within 2 points of Opus 5.5 on GDPval-AA. It also generates output 30%+ faster than Sonnet 5 and keeps the same $2/$10 per million token price. Anthropic says cost per task falls by up to 30% because the model uses fewer tokens. You can deploy it today via the Claude API, AWS, Google Cloud and Azure.
NVIDIA Launches Open Agent Safety Platform: OpenShell Sandboxes Agents on Vera CPUs While Sentry on BlueField-4 Quarantines Them in Milliseconds
NVIDIA has launched the Open Agent Safety Platform, an open reference design that enforces AI agent safety outside the agent itself. OpenShell, an Apache 2.0 runtime, sandboxes agents under YAML policies. Sentry, an out-of-band watchdog on BlueField-4 DPUs, can quarantine an agent that escapes its boundary in milliseconds. NVIDIA says over 100 organizations are working with the platform.