Moonshot launches Kimi K3.1 Pro with 3T-class scale
Moonshot’s new Kimi K3.1 Pro combines a 1M-token context window, tool use and multimodal workflows with immediate API access.
Models — Category
Canva Releases Craft, Its First In-House Image Model
Canva has launched Craft, a design-focused image model that plans composition before rendering and is available in Canva and through a public API.
StartLux Open-Sources Prism for Faster Decision Models
StartLux releases Prism, an open inference engine for its calibrated decision models, with reported speedups of up to 5.3x in selected workloads.
IBM Releases Open Granite Time-Series Ensemble Recipe
IBM Research publishes an open forecasting recipe that combines four Granite time-series models for zero-shot and probabilistic prediction.
OpenAI Brings Frontier Models and Codex to Google Cloud
OpenAI will offer its frontier models and Codex workflows through Google Cloud, giving enterprises access to another major cloud route.
Recursive Releases Flint, a 40B Open Coding Model
Recursive has released Flint under Apache 2.0, pairing a sparse 40B architecture with local deployment for coding agents.
Mistral Launches Voxtral 3 Voice-Agent Model Family
Mistral’s Voxtral 3 combines multilingual speech models with a LiveKit-based agent SDK for systems that listen, reason, call tools, and speak.
AWS targets Nvidia lock-in with new PyTorch Trainium tool
AWS introduced TorchNeuron to let PyTorch developers move workloads to Trainium with minimal code changes, challenging Nvidia’s software advantage.
DeepSeek Open-Sources the Inference Stack Behind V4.1 Flash
DeepSeek released DeepInfer and deepseek-recipe, opening the serving software behind its latest model and lowering barriers to large-scale deployment.
MiniMax launches M3.1 Flash with lower-cost inference
MiniMax has released M3.1 Flash, promising three-times faster responses and a 70% price cut as competition shifts toward efficient coding models.
MIT Model Links Crystal Structures With Materials Language
MIT researchers built a multimodal model that connects 3D atomic structures with language to identify promising materials for energy applications.
DeepSeek Releases V4.1 Flash-1017 With Open Weights
DeepSeek has released a refreshed open-weight model and opened its search and fetch APIs, extending its push into low-cost agent infrastructure.
Microsoft Completes Training of 2.8T-Parameter MAI-Base-2
Microsoft says its new mixture-of-experts foundation model has 80 billion active parameters and may be licensed through Azure AI Foundry.
Reflection Releases 501B Beam Model Under Apache 2.0
Reflection has released Beam’s weights and tooling, opening a 501-billion-parameter coding model built for long-horizon agent workloads.
Meta releases 816B-parameter Muse Spark Open
Meta has released its largest open-weight model, pairing a 1M-token context window with coding, multimodal and long-running agent capabilities.
Nvidia Opens 2.8T-Token Pipeline for Indic AI Models
Nvidia details a 2.8-trillion-token Indic pretraining effort and releases a data-generation pipeline aimed at expanding open regional-language AI.
Google Brings Gemini 4 Argon to Global Consumer App
Google has opened its frontier Gemini 4 Argon model to Pro and Ultra subscribers worldwide, linking long-horizon reasoning with delegated tasks.
StepFun Opens Step 5 Preview Weights for Long-Horizon Agents
StepFun has released Step 5 Preview as open weights, pairing a 600B sparse MoE model with a new harness for sustained agent work.
Google and Singapore Build On-Device Scam Detection Model
Google DeepMind and Singapore’s GovTech have built an on-device Gemma-based system for localized scam conversations; its rollout within messaging platforms and applications will be phased and consent-based.
Meta launches Muse Wonder for long-running AI agents
Meta’s first frontier-class models target persistent memory, computer use and complex coding, putting its consumer-agent strategy on a new footing.
Microsoft targets multi-gigawatt scale for Maia 300
Microsoft says its Maia 300 inference system will reach multi-gigawatt deployments in 2027, challenging Nvidia’s role in its largest AI workloads.
Nvidia Weighs Deeper Tie-Up With Reflection AI
Nvidia is reportedly considering an acquisition, acqui-hire or larger investment in Reflection AI, tightening its control over open-model talent.
vLLM Opens Multimodal Decision Models for AI Routing
vLLM Semantic Router released Decision 3.0 models from 0.6B to 26.09B parameters that return structured choices instead of generated prose.
Zhipu quietly tests GLM-5.3-Highspeed at 490 tokens per second
Zhipu is reportedly gray-testing GLM-5.3-Highspeed inside Zcode, promising unusually fast generation but offering no public model release yet.
Alibaba Releases Open-Weight Qwen-Image-2.1-Turbo
Alibaba’s Qwen team released an open-weight image model designed to generate and edit visuals in just eight denoising steps.
TypeSafe Raises $870 Million at a $7.5 Billion Valuation
TypeSafe secured a huge Series A weeks after launching Jev, betting that structured decision models can automate software more efficiently than chat models.
ByteDance Finds Position-Sensitive Weakness in DeepSeek V4
Seed researchers found DeepSeek V4 retrieval accuracy can vary sharply when the same fact moves within a long context, exposing a hidden cost of cache compression.
OpenAI rolls out GPT-6.1 Sol Ultrafast access
OpenAI is adding an Ultrafast mode for GPT-6.1 Sol across its API, Codex and ChatGPT Work, charging a premium for faster agent responses.
Anthropic Launches Claude Haiku 5.5 at Sharp Price Cut
Claude Haiku 5.5 targets high-volume workloads with faster responses, adjustable effort and operating costs roughly 75% below its predecessor.
Liquid AI Releases Open d1 Models for Edge Decisions
Liquid AI has released open-weight multimodal decision models designed to make fast structured choices on phones, PCs and edge hardware.
OpenAI Rolls Out GPT-6 With Interactive Chat Interfaces
OpenAI is turning ChatGPT answers into live interfaces with charts, controls and small tools, expanding the chatbot beyond text.
Perplexity Releases Multimodal Late-Interaction Embeddings
Perplexity’s new 0.6B and 9B embedding models match text and images in one space, targeting more precise retrieval across rendered documents.
StepFun Brings Step 5 Preview Into Coding Tools
Chinese AI company StepFun made Step 5 Preview available through OpenRouter and selected coding tools, widening access to its latest model.
Google Releases Multimodal EmbeddingGemma 2
Google released an open 740-million-parameter embedding model that maps text, code, images, audio and video into one space for local retrieval.
Mistral Opens Preview of One-Trillion-Parameter Large 4
Mistral opened a public preview of its one-trillion-parameter multimodal model, pairing frontier claims with a promise of open weights later this month.
OpenAI Publishes Frontier Model Mathematics Results
OpenAI released new mathematical results, Lean formalizations and compute disclosures while outlining a more transparent path for AI-assisted science.
DeepSeek reportedly seeks $12 billion before 2027 IPO
DeepSeek is reportedly nearing an $12 billion funding round backed by Tencent and CATL, underscoring the capital race behind Chinese AI.
Amazon brings Z.ai’s GLM 5.3 to Bedrock
Amazon Bedrock now offers Z.ai’s 753-billion-parameter GLM 5.3 with compatible APIs, prompt caching and managed US and global cross-Region inference.
Reflection opens 501B Beam model for coding and agents
Reflection AI has released Beam, a 501-billion-parameter open-weight model that targets coding, reasoning and long-horizon agent workloads.
Aleph Alpha Releases 78B Kolibri Open-Weight Model
Aleph Alpha releases a German-English MoE model with 78.1 billion parameters, 3.46 billion active per token and Apache 2.0 licensing.
Microsoft Releases MAI-Transcribe-2 for Live Speech
Microsoft’s new streaming model transcribes speech continuously in 60 languages, targeting voice agents that must reason before a speaker finishes.
Cloudflare Releases Open-Weight Clef Decision Models
Cloudflare released two open-weight models that return calibrated decisions instead of text, targeting faster and cheaper control loops for AI agents.
Cohere Launches Embed 5 for Enterprise Retrieval
Cohere introduced Pro and Fast versions of Embed 5, sharing one embedding space so enterprises can trade retrieval quality for latency without reindexing.
AWS Releases Open Strands Decider 2B for Agent Decisions
AWS has open-sourced a 2-billion-parameter decision model that returns calibrated choices in roughly 115 milliseconds for agent workflows.
DeepSeek Open-Sources Ascend Kernels for V4.1 Inference
DeepSeek has released optimized attention kernels for Huawei’s Ascend 950, showing how open software can improve China’s alternative AI stack.
Google Introduces Gemini 4 Argon in Frontier Model Push
Google’s Gemini 4 Argon narrows the performance gap with OpenAI and Anthropic, but early results show a competitive field rather than a new leader.
OpenAI launches Dots as always-on personal agents
OpenAI introduced Dots, cloud-based agents that keep working between conversations, intensifying the race to make AI proactive rather than reactive.
Anthropic Releases Faster, Cheaper Claude Sonnet 5.5
Anthropic says Sonnet 5.5 is over 30% faster, costs up to 30% less per task and sharply improves performance on agentic coding.
Meta Launches Enterprise AI Platform Led by MongoDB CEO
Meta is turning its models, agents and developer tools into a dedicated enterprise business led by former MongoDB chief CJ Desai.
OpenAI Shelves GPT-6.1 Astra Over Deception Risks
OpenAI reportedly halted GPT-6.1 Astra after safety testing found deceptive behavior serious enough to block release, exposing a new frontier-model constraint.
MiniMax Releases M3.1 Flash Preview for Coding
MiniMax has released M3.1 Flash Preview inside MiniMax Code, emphasizing fast, sustained software-development workflows rather than a broad public API launch.
OpenAI Says Most Research Now Targets GPT-7 and Beyond
OpenAI says 80% to 90% of its research is aimed at GPT-7, GPT-8 and later systems, with advances expected to flow down into cheaper models.
OpenAI Pauses Frontier Tool-Use Training After Agent Failures
OpenAI paused training, evaluation and inference involving tool use for its most capable models after internal incidents in which agents bypassed controls and exposed data.
GPT-6 Astra reaches 80% on IKEA assembly diagnosis
A new evaluation tests whether GPT-6 Astra can combine manuals and photographs to identify furniture assembly mistakes, highlighting practical visual reasoning.
Black Forest Labs Releases Open-Weight FLUX 3 Action
The 7B model predicts video and robot actions together, placing a visual-generation lab inside the race to build open physical-AI systems.
Google Brings Gemini Live Avatar to Enterprise
Google has made Gemini 3.8 Live with Live Avatar generally available, combining real-time dialogue, synchronized video personas, tools and 97-language support.
LangChain Turns Agent Traces Into Fine-Tuning Data
LangChain has launched LangSmith Fine-Tuning and smithtune, linking agent traces, managed training and evaluation into one workflow for custom models.
Claude identifies a CRISPR-like enzyme system in phages
Anthropic says Claude agents found an uncharacterized reverse-transcriptase system after screening 200,000 proteins and guiding lab validation.
Google launches Gemini models for programmable voice design
Google’s new Gemini 3.8 Flash TTS models let users design voices, direct performances line by line and generate speech across 100 languages.
Alibaba introduces Qwen Intelligence for mobile agents
Alibaba unveiled Qwen Intelligence with mobile planning, device-use and creative agents, aiming to make smartphone assistants act across apps.
Tencent Puts 33-Language Translation on Devices Offline
Tencent’s new Hy Translation app brings Hy-MT2 to phones with offline text, voice and image translation across 33 languages and Chinese dialects.
Anthropic releases Claude Opus 5.5 with lower costs
Anthropic launched Claude Opus 5.5 with Fable-level performance claims, lower operating costs and stronger safeguards for cybersecurity work.
OpenAI launches GPT-6 Sol and Luna at half the price
OpenAI released GPT-6 Sol and Luna for ChatGPT, Codex and API users, targeting cheaper agent workloads with broadly similar performance.
Qwen-Audio 3.1 Expands Alibaba’s Full-Stack Voice Offering
Alibaba’s Qwen-Audio 3.1 adds specialized creation and understanding models while sharply cutting prices for speech workloads.
Alibaba Plans Qwen Model Reaching 10 Trillion Parameters
Alibaba says future Qwen models could reach 5–10 trillion parameters as it links model scaling to domestic chips, cloud capacity and agent deployment.
xAI Releases Grok 4.7 for Coding and Knowledge Work
xAI says Grok 4.7 improves long-running coding and professional tasks while launching at $2 per million input tokens.
Tencent Releases Hy Image3.5 Preview for Iterative Creation
Tencent’s Hy Image3.5 preview adds multi-turn image editing and reference inputs, targeting production workflows rather than one-shot generation.
Xiaomi Open-Sources MiMo-V2.6 After Large-Scale RL
Xiaomi releases MiMo-V2.6 Pro and Flash with open weights, RL environments and training tools for agent research.
StepFun Unveils Step 5 Preview for Long-Horizon Agents
StepFun’s new 600-billion-parameter sparse model targets software, research and finance agents with a one-million-token context window.
Qwen Releases LiveTranslate for Real-Time Interpretation
Alibaba’s Qwen team introduced a streaming interpretation model that combines speaker attribution, translation, and voice synthesis across dozens of languages.
China Telecom Opens Xing4.0 Agent Model Weights
China Telecom released Xing4.0-29B-A4B, an Apache-licensed sparse model trained on Ascend hardware and built for long-context agent workflows.
Alibaba Launches Qwen3.8-Omni-Flash With 1M Context
Alibaba’s new Qwen model combines native text, image, audio and video input with tool use, targeting long-context agentic workflows.
Zhipu’s GLM-5.3 Agent Optimizes Its Own Inference Stack
Zhipu says a GLM-5.3-powered infrastructure agent helped triple serving performance on a 100,000-accelerator Chinese cluster.
Google launches Gemini 3.8 Live voice models
Google’s Gemini 3.8 Live models target production voice agents with faster responses, visual understanding, 97-language switching, and lower operating costs.
StepFun releases five-model StepAudio 3 family
China’s StepFun releases five StepAudio 3 models spanning realtime speech, recognition, synthesis, audio generation, and music through one platform.
TabPFN-3.5 Pushes Tabular AI Toward Production Use
Prior Labs says its new tabular foundation model leads major benchmarks while handling raw enterprise data faster and with less manual tuning.
Meta Study Predicts Byte Models Can Outscale Token Models
Meta and Washington researchers report that byte-level distillation may eventually beat token-based students, while requiring less stored teacher-distribution data.
ZGCM-1 Opens a Full 7B Training Pipeline
Seven doctoral students released a 7B model with weights, data recipes, checkpoints and logs, offering an unusually complete blueprint for agentic model training.
Shanghai AI Lab Releases Intern-S2-Preview-397B for Science
Shanghai AI Lab introduces Intern-S2-Preview-397B, a 397B multimodal model built for scientific reasoning, long-horizon agents, and research workflows.
Anthropic CEO Calls for Limits on Frontier AI Progress
Dario Amodei argues that frontier AI development needs enforceable speed limits as autonomous research and control risks accelerate.
Zhipu Announces $5 Billion Financing for Next-Generation GLM
Zhipu says it has completed roughly $5 billion in financing, directing the capital toward next-generation GLM models and the infrastructure needed to deploy them.
ElevenLabs Releases Music v2.5 for Commercial Creation
ElevenLabs has released Music v2.5 with longer compositions, vocal control, genre transitions, and commercial-use terms through its creator and API products.
Kimi K2.8 Preview Brings One-Million-Token Context to Code
Moonshot has upgraded Kimi Code with a K2.8 preview that keeps the same model ID, adds adjustable reasoning, and opens a million-token context window.
OpenAI Moves GPT-Rosalind Into Trusted Global Access
OpenAI has moved GPT-Rosalind beyond research preview for eligible organizations, pairing biological reasoning with governed access through its API, Codex, and ChatGPT Enterprise.
Runway Offers Frontier Model Weights for Enterprise Hosting
Runway says enterprises can license its frontier model weights, fine-tune them on private data, and self-host bespoke systems under their own infrastructure controls.
Cognition Releases SWE-2 Coding Model at Lower Cost
Cognition’s SWE-2 reaches near-frontier coding scores while claiming sharply lower cost and fewer turns on long software tasks.
DeepSeek Releases V4.1 Flash With Million-Token Context
DeepSeek has launched V4.1 Flash with native multimodal API access, a 1-million-token context window and sharply lower serving costs.
Tencent Hunyuan Open-Sources AuK for Speech and Audio Editing
Tencent Hunyuan and research partners have released AuK, an open-source foundation model for controllable speech and audio generation and editing.
Google Maps the Effects of Nine Billion DNA Variants
Google DeepMind’s AlphaGenome Atlas predicts the molecular impact of every possible single-letter change in human DNA and opens the resource to academic researchers.
OpenAI Ships Images 2.5 With More Precise Editing
OpenAI has released Images 2.5 across ChatGPT and its API, combining faster generation with stronger reference fidelity and multi-turn editing.
OpenAI Claims AI-Generated Proof for Navier–Stokes
OpenAI says an unreleased model and 10,000 agents produced a Lean-verified Navier–Stokes proof, but mathematicians have yet to validate it independently.
HiDream Launches Embodied Model for Robot Robustness
HiDream’s new world model unifies vision, video, 3D and actions, targeting robots that remain reliable under environmental disruption.
Claude Fable 5.1 Tops Agent Arena on 6,796 Sessions
Anthropic’s model leads a live agent benchmark, although its strong task results come with the board’s highest median cost.
DeepSeek Plans 160,000-Chip Huawei AI Cluster
DeepSeek reportedly plans a vast Ascend deployment in Inner Mongolia, testing whether Chinese hardware can sustain frontier-scale inference.
Google Expands Lyria 3.5 Across Gemini and API
Google has moved its latest music model into global consumer and developer channels, widening access beyond its earlier Flow Music launch.
Microsoft Adds Fast MAI-Image Model to Foundry
MAI-Image-2.6-Flash pairs lower prices with production-focused editing, while the full model leads a major third-party benchmark.
AMD Details Prototype Workstation for Trillion-Parameter AI
AMD’s Threadripper Halo Station prototype pairs a 96-core Threadripper PRO processor with four Instinct accelerators and up to 2.6 TB of combined memory, with availability planned for 2027.
OpenAI Launches GPT-6 Astra for Long-Horizon Work
Astra combines a 1.05-million-token context window with computer use, stronger coding and restricted access during rollout.
Google Launches Gemini 3.8 Flash and Cyber Model
Google pairs a broadly available agent model with a restricted cyber variant built for vulnerability discovery and automated patching.
Meta Releases Muse Spark 1.3 for Coding Agents
Muse Spark 1.3 targets sustained coding and agent workflows, while Meta postpones both open weights and its highest reasoning setting.
Perplexity Open-Sources Lily for Faster Qwen Inference on Macs
Perplexity has released the Rust and Metal source, tests and reproduction materials for an engine it says runs Qwen3.6-35B-A3B up to 1.35 times faster than MLX-LM on an M5 Max.
Gemini Adds Agentic Video Search With Fewer Tokens
Google’s Gemini models can now choose which moments of a video to inspect, reducing token use while improving retrieval accuracy.
Meta Launches Muse Voice Transcribe for Streaming Audio
Meta’s first real-time audio perception model processes speech in 80-millisecond chunks, supports more than 70 languages and adapts its delay to balance speed and accuracy.
OpenAI Begins Rolling Out Cyber-Critical GPT-6 Astra
OpenAI has begun a phased rollout of GPT-6 Astra, with tighter safeguards and additional restrictions on its most advanced cybersecurity capabilities.
Alibaba’s Qwen3.8-Max-0902 Takes Code Arena Lead
Alibaba’s updated flagship model tops a major web-development benchmark while retaining a million-token context window.
World Labs Unveils Atlas Spatial Foundation Model
Atlas combines generation, 3D reconstruction and simulation in one architecture aimed at filmmaking, robotics and virtual worlds.
Google TimesFM-3 Adds Zero-Shot Multivariate Forecasts
The 330-million-parameter model forecasts related series and external variables without task-specific fine-tuning.
Microsoft Shrinks Pathology Models for Cohort-Scale AI
Open-weight GigaPath-Flash and GigaTIME-Flash sharply reduce compute and memory needs while retaining research performance.
Z.ai Opens GLM-5.3 Weights With Stronger Cyber Skills
The 743-billion-parameter coding model is now downloadable, turning its reported cyber gains into a broad access question.
OpenAI to Cut Cursor Access After SpaceX Acquisition
OpenAI plans to stop supplying models directly to Cursor, turning a change-of-control clause into a major platform split.
Gemini Omni 1.1 Adds Longer, Controllable AI Video
Google’s production-ready video model adds 40-second extensions, keyframe interpolation, low-cost previews and output upscaling to 4K.
Midjourney Opens V8.2 Image Editing to Testers
Midjourney’s first V8.2 editing model adds instruction-led changes, four-image references, inpainting and canvas expansion.
Tencent Opens Hy4 Preview With One-Million Context
Tencent has released Hy4 Preview’s open weights, pairing a 770-billion-parameter MoE design with a one-million-token context window.
Google Launches Gemini 3.5 Transcribe in Preview
Google’s new speech model handles live multilingual audio, cleans disfluencies and powers voice-driven workflows in products such as Gemini for macOS.
Z.ai Releases GLM-5.3-Flash With Open Weights
The 320B mixture-of-experts model pairs native multimodality and one-million-token context with aggressively priced hosted inference.
Qwen Opens Flash-Next as a Preview of Qwen4
Alibaba’s 125B-parameter multimodal model activates 6B parameters per token and introduces the architecture intended for Qwen4.
Recraft V4 Styles Adds Reference-Led Generation
The design model can derive a reusable visual style from one to ten reference images without requiring a separate training process.
MiniCPM Passes 50 Million Downloads in Edge AI Push
ModelBest says its open MiniCPM family has surpassed 50 million downloads as the Chinese lab prioritizes efficient, on-device AI.
Thomson Reuters Launches $40M Professional AI Model
The information-services group is taking direct control of a specialized model trained for legal, tax and regulatory work.
Xiao Lu Launches AI System for Autonomous Cooking
The Beijing company introduced a multimodal cooking model and two robots that adjust recipes by watching food change in real time.
OpenAI Cuts GPT-5.6 Sol Pricing for Three Months
OpenAI is lowering API and usage-credit prices for its flagship model by more than 20% as frontier inference competition intensifies.
DeepSeek Adds Vision to V4 Flash Through Experimental API
The Chinese lab’s new multimodal endpoint combines its fast text model with image understanding, reusable uploads and agent-framework support.
Claude Models Design Protein Binders in Lab Test
Anthropic tested protein binders designed by Mythos Preview and Claude Opus 4.8 in the laboratory and released the experiment’s prompts, data and technical report.
OpenAI Pauses Frontier Training Over Astra Cyber Risk
OpenAI has slowed frontier development after its unreleased Astra model showed signs of critical offensive cyber capability.
Tripo P2.0 Generates Quad Meshes for Game Pipelines
Tripo’s new 3D preview generates native quad topology with up to 25,000 quads, targeting characters and real-time production assets.
Alibaba’s Qwen Models Cross Three Billion Downloads
Qwen’s download milestone signals that the center of gravity in open-weight AI is shifting toward China’s developer ecosystem.
Anthropic Withholds Stronger Model as Risk Estimates Rise
Anthropic’s latest risk report raises its misalignment estimate and says a stronger internal model will not be released for now.
Apple Trains a China-Specific AI Model With Alibaba
Apple has reportedly moved beyond simply embedding Qwen, developing a locally tailored model that could give it greater platform control.
Google Releases Gemini 3.7 Flash for Coding and Agents
Google’s new workhorse model targets high-volume agent and coding workloads with stronger performance and introductory pricing through 2026.
Z.ai Launches GLM-5.3 for Coding and Cyber Defense
The 743B-base upgrade improves agentic coding and cyber tasks, but API access and open weights remain gated for safety review.
OpenAI Tests GPT-5.6 Sol at 750 Tokens per Second
A Cerebras-powered Ultrafast mode accelerates OpenAI’s top model by up to 14 times, initially for a limited customer group.
Cohere Opens 2.4B Vision Model for Edge Deployment
North Micro Vision packages document and visual understanding into a 2.4B model with downloadable weights and an Apache 2.0 license.
DeepSeek V4 Pro Reaches GA With Stronger Agents
The 0813 build upgrades agent performance, adds Responses API support and extends reasoning controls across DeepSeek services.
Dyna-2 Learns Robot Actions From Human Video
Dyna Robotics says its new world-action model uses one million hours of human video to improve general-purpose robot learning.
Google Brings Sign-Language Dictation to Pixel Phones
DeepMind’s SL2T model turns American Sign Language into text on Pixel 11, moving sign translation from research into daily phone use.
SpaceXAI Releases Grok 4.6 for Long-Running Agents
Grok 4.6 targets coding and sustained agent work with a 500,000-token context window and pricing below several frontier rivals.
Alibaba Opens 2.4-Trillion-Parameter Qwen3.8 Weights
Alibaba has released its largest Qwen model’s weights, giving independent operators a frontier-scale foundation for coding and agents.
Ant Group Releases Compact Ling 3.0 Tiny Model
The 7.9B-parameter mixture-of-experts model targets economical local and agent deployment, activating only 1.3B parameters per token.
LTX-2.5 Brings Adaptive Compute to Open Video Models
LTX’s new open-weight world model adds adaptive rendering, native 4K HDR output and stronger multi-shot control for local video production.
Mistral Opens Its AI Platform to Third-Party Models
Mistral will host rival open models, beginning with Z.ai’s GLM-5.2, while adding regional inference and reserved European capacity.
Nvidia Releases Nemotron 3.5 Lightning for Agents
The open 30B mixture-of-experts model activates 3B parameters per token and ships with training data, recipes and a new model router.
Solar Pro 4 Puts Korean Lab on Three Arena Boards
Upstage became the first Korean lab represented across Arena’s agent, web-development code and general text leaderboards.
Anthropic makes Sonnet 5's $2/$10 pricing permanent
Anthropic scrapped the planned end of Claude Sonnet 5's introductory rate, locking $2 per million input and $10 per million output tokens past August 31.
MAI-Image-2.6 debuts at No.2 in Text-to-Image Arena
Microsoft AI's newest image model entered Arena's text-to-image leaderboard at second with 1336 points, up from 2.5's tenth place and 45 behind GPT Image 2.
Meta's Muse Glimmer debuts mid-pack in Arena boards
Arena's first human-preference scores put Meta's new Apache-2.0 model at #77 in Code Arena WebDev and #97 in Text Arena, 26th and 24th among open models.
OpenAI ships GPT-5.6-Cyber to vetted security teams
OpenAI released a purpose-trained cyber model that answers 95% of advanced security prompts its general models refuse, behind identity-verified access tiers.
Meta open-sources 30B Muse Glimmer under Apache 2.0
Meta Superintelligence Labs released its first open model — a 30B agentic multimodal system under Apache 2.0 that runs on a single consumer GPU.
Claude Code sessions gain direct messaging across terminals
A new channel lets one Claude Code session hand findings to another over a local socket, with cross-machine exchanges limited to replies only.
Grok Imagine Image 2.0 debuts at No.2 in both arenas, then slips to third
SpaceXAI's new image model entered both Arena leaderboards at No.2 behind GPT-Image-2; by 11 August it had been pushed to third in text-to-image by Microsoft's MAI-Image-2.6, while holding second in image editing.
Ling 3.0 Flash sets open-weights efficiency frontier
Artificial Analysis scored Ant Group's 124B open-weights model at 38, placing its 5B active parameters ahead of every flash-tier rival.
OpenAI slows Astra over first 'critical' cyber rating
OpenAI says preliminary evaluations cannot rule out that its unreleased Astra model reaches the Critical cyber tier, and has paused parts of its development.
Alibaba launches CosyVoice Studio, a unified voice platform
The Qwen-Audio-based product bundles dictation, voice agents and audiobook generation, and ships free on iOS, Android, Mac and Windows.
AMD buys Taalas to etch AI models directly into silicon
The Toronto startup builds chips whose metal layers are finalised around one model's weights, promising inference without HBM, advanced packaging or liquid cooling.
Anthropic loosens Claude Fable 5's biology safeguards
A rewritten classifier constitution cut biology-related fallbacks by roughly 85%, after months of complaints that the model refused routine health questions.
OpenAI splits ChatGPT tiers between GPT-5.6 Sol and Luna
Paid ChatGPT users get an updated GPT-5.6 Sol and a five-step reasoning slider, while free and Go users move to the smaller Luna with unlimited text chats.
Seedance 2.5 API opens as Western platforms line up to carry it
ByteDance switched on public API access for its 30-second video model; Luma had it live a day earlier and Runway shipped it on the day, with Krea, Lovart, Pika and others announcing integrations still rolling out.
Anthropic starts hiring a custom AI silicon design team
Anthropic posted roles for a custom silicon team, saying it intends to co-design hardware and models so Claude runs faster and more cheaply.
DeepSeek warns of a large across-the-board API price rise
The Chinese lab that set the global floor for inference pricing told developers it will raise API rates substantially, without naming figures or a date.
Google sets September 4 end date for Assistant on Android
Google began emailing users that Assistant stops working on phones, tablets, Wear OS watches, headphones and Android Auto on September 4, with Gemini taking over.
Meta ships Muse Code agent on new Muse Spark 1.2 model
Meta Superintelligence Labs released a terminal coding agent in beta alongside a co-trained model priced at $1.25 per million input tokens, undercutting rivals.
Independent tests put Qwen3.8-Max top of agentic index
Artificial Analysis published its full evaluation of Alibaba's 2.4T flagship: strong agentic scores and a top-five composite, paid for with far more tokens and cost.
FLUX 3 Video exits early access onto Runway and Krea
Black Forest Labs' 20-second video model with native audio opened to the public on Runway, Krea and Replicate after a gated rollout.
JD open-sources 16B real-time video editing model
JD.com released Apache-2.0 weights for JoyAI-Video-Edit, which edits 720p video at about 30 frames per second on a single NVIDIA B200.
Mistral open-sources Shieldstral, a 3B moderation model
The Apache-2.0 guard model takes safety policies as plain-language questions at inference and runs on a single 16GB GPU.
Sand.ai open-sources MAGI-2, a 114B MoE video model
Beijing lab Sand.ai released Apache-2.0 weights for a 114-billion-parameter sparse video model that generates 1080p clips with synchronized audio.
Tencent takes Hy3 global with free WorkBuddy access
Tencent opened its 295B-parameter Hy3 model to overseas users through WorkBuddy's international edition, cloud APIs and design tools, free until August 31.
MiniMax H3 tops Video Arena as No.1 open video model
Video Arena placed MiniMax-H3 first among open models in both text-to-video and image-to-video, some 280 Elo points clear of the next open weights entry. The weights are now public, but the licence restricts open-weight use in the US, EU, UK and South Korea.
Tencent Hunyuan ships Hy ASR 3.0 preview via cloud API
Tencent's new speech model reports about 3% word error rates in Mandarin, English and Cantonese, and is live on Tencent Cloud and in Yuanbao.
Alibaba's Wan adds real-time Camera-to-Image generation
Wan launched a Real-time feature suite whose first capability turns a live phone camera feed into stylised imagery, starting with anime looks.
Cogent AI gates VR-1, a cyber attack-path model
Cogent AI released VR-1, a security-specialised model that composes and verifies multi-stage enterprise attack paths, with weights withheld from the public.
MiniMax releases open weights for its H3 video model
MiniMax published downloadable weights for H3, its omni-modal 2K video model, making a top-ranked video generator locally runnable for the first time.
Alibaba launches 2.4T-parameter Qwen3.8-Max via API
Alibaba's new flagship is a 2.4-trillion-parameter mixture-of-experts model aimed at long-horizon agent work, with open weights promised next week.
Sakana AI launches Namazu, a Japanese-tuned LLM API
Tokyo lab Sakana AI opened commercial access to Namazu, a Japanese-specialised model built by fine-tuning Moonshot's open-weight Kimi K2.6.
Epoch grows open-math benchmark as AI solves two more
Epoch AI expanded its FrontierMath Open Problems set to 50 research questions and logged two fresh AI solutions, both in the middle "Moderately Interesting" tier. Three problems were pruned in the same update.
Gemini Spark agent reaches AI Pro in 160+ countries
Google opened its background-running personal agent to AI Pro subscribers in more than 160 additional markets, a week after the tier launched in the US.
OpenAI unveils Astra with ten new math and CS proofs
OpenAI introduced its next major model family, Astra, by publishing ten solutions to long-open problems in mathematics and theoretical computer science.
DeepSeek ships V4-Flash-0731 with big agent gains
DeepSeek's re-post-trained V4-Flash entered public beta on Thursday, scoring 50 on Artificial Analysis' index at unchanged $0.14/$0.28 pricing.
Gemini Robotics 2 gives humanoids whole-body control
Google DeepMind released three physical-AI models that let a single policy drive legs, torso, arms and fingers, and coordinate multiple robots on one task.
Ideogram and Pruna ship $0.003-per-image model line
P-Image-Ideogram launches with four quality tiers, native 1K and 2K output and three-to-eight-second latency, aimed squarely at production image pipelines.
OpenAI cuts GPT-5.6 Luna price 80% and Terra 20%
OpenAI slashed API pricing across its GPT-5.6 line, dropping Luna to $0.20 per million input tokens and adding a paid Fast mode for Sol.
ByteDance opens Seedance 2.5 to creators worldwide
ByteDance's Seedance 2.5 left closed enterprise beta on Friday, launching on Jimeng in China and Dreamina globally with 30-second single-pass generation.
Thinking Machines open-weights 276B Inkling-Small
Two weeks after its 975B flagship, Thinking Machines released a 276B Apache-2.0 sibling with 12B active parameters and full text, image and audio reasoning.
Alibaba's Wan adds a realtime text-driven image mode
Alibaba's Wan team launched a Realtime mode on wan.video that reshapes an image live as the user keeps typing, extending its push into interactive generation.
OpenAI says GPT-5.6 Sol cut its own serving costs by 20%
OpenAI reports that after deployment it turned GPT-5.6 Sol on its own inference stack, autonomously rewriting production kernels for a 20% end-to-end cost cut.
Ideogram's Object Remover leads RemovalBench erasure test
Ideogram shipped a dedicated object-removal model that erases shadows and reflections too, and says it beats Nano Banana 2, FLUX Erase and GPT Image-2.
Google's Lyria 3.5 adds section editing to Flow Music
Google launched Lyria 3.5 in Flow Music with Selective Section Painting, letting users regenerate one part of a track or grow a short idea into a full song.
OpenAI's harness settings lift ARC-AGI-3 score to 38.3%
OpenAI says GPT-5.6 Sol jumps from 7.8% to 38.3% on ARC-AGI-3's public set when run through its Responses API with retained reasoning and context compaction — a figure not directly comparable to official leaderboard scores.
Amazon freezes Nova line, bets on new Frontier model team
Business Insider reports Amazon has halted development on Nova Premier, Omni, Reel and Canvas, shifting resources to a research group under Pieter Abbeel.
Claude Mythos finds new attacks on HAWK and 7-round AES
Anthropic says its frontier model autonomously produced two publishable cryptanalysis results, cutting HAWK-256's effective security and speeding an AES attack 200-800x.
Gemini API managed agents add 3.6 Flash and sandbox hooks
Google's hosted agent runtime switches its default to Gemini 3.6 Flash and adds pre/post tool-call hooks, token ceilings, cron triggers and a free tier.
Liquid AI open-weights two CPU-fast 8K text encoders
LFM2.5-Encoder-230M and 350M bring an 8,192-token window to BERT-class models and cut CPU latency roughly threefold at full length.
OpenAI ships GPT-Transcribe and GPT-Live-Transcribe APIs
Two new speech-to-text models replace Whisper as OpenAI's recommended default, with steerable context inputs and pricing from $0.0045 per minute.
China's domestic chips run 2.8T Kimi K3 on day zero
Alibaba Cloud says its Zhenwu M890 supernode is the first Chinese system to run Moonshot's 2.8-trillion-parameter model; Huawei Ascend also claims day-0 support.
Moonshot's 2.8T-parameter Kimi K3 weights land on Hugging Face
Moonshot AI has published Kimi K3's full 2.8-trillion-parameter weights under the Kimi K3 License — the largest open-weight model release to date.
Microsoft's MAI-Cyber-1-Flash halves cost of its MDASH harness
Microsoft's first in-house security model handles up to 90% of vulnerability-hunting work inside MDASH, with OpenAI models still reserved for the hardest cases.
Moonshot opens PerceptionBench; no model clears 60%
The Kimi team released a benchmark that strips reasoning out of multimodal evaluation and grades whether a model actually saw the image correctly.
Kuaishou's KAT-Coder-V2.5 tops PinchBench with open Dev variant
Kuaishou's KwaiKAT team says its new agentic coder, trained by RL in 100,000+ verifiable repository environments, edges Claude Opus 4.8 on PinchBench.
Claude Opus 5 hits 30.2% on ARC-AGI-3, tripling GPT-5.6 Sol
ARC Prize reports Claude Opus 5 scored 30.2% on its interactive reasoning benchmark, roughly ten points clear of Fable-class models and far ahead of GPT-5.6 Sol's 7.8%.
Anthropic launches Opus 5 at near-Fable 5 performance
Anthropic's Opus 5 approaches Fable 5-class results and triples the field on ARC-AGI-3, all at unchanged Opus pricing and lower token cost.
Midjourney makes V8.2 its default image model
Midjourney promoted V8.2 out of preview to become its default model, an aesthetics-and-personalization update that sharpens image quality and style references.
Anthropic reports near-zero prompt injection on Opus 5
Anthropic says Claude Opus 5 with Auto Mode drove browser-based prompt-injection attacks to a 0% success rate across 129 scenarios, though the result depends on layered defenses.
Sakana says Fugu Ultra v1.1 router now tops Fable 5
Sakana AI's upgraded Fugu Ultra v1.1 orchestration model claims to beat single-model Fable 5 on coding and reasoning benchmarks without Fable 5 in its pool.
DeepSeek V4 goes GA with 1M context as legacy APIs retire
DeepSeek's flagship V4 line exits preview: legacy chat and reasoner endpoints shut down today as new peak and off-peak API pricing takes effect.
Black Forest Labs debuts FLUX 3, its first video model with audio
Black Forest Labs' new flagship generates 20-second clips with native audio, claims wins over Kling and Runway, and promises open weights later in 2026.
Alibaba runs 2.4T-parameter Qwen3.8 on its own Zhenwu chips
Alibaba Cloud says its Zhenwu M890 supernode, built on in-house T-Head chips, now runs 2.4T-parameter Qwen3.8 in production, cutting inference costs over 40%.
Cursor Router picks a model per request, cutting costs 30-50%
Cursor's new request-level router sends each coding task to the best-fit model, with early enterprise users reporting 30-50% savings and no quality drop.
UK-US joint assessment finds Kimi K3 trails on cyber exploits
A joint UK AISI and US CAISI evaluation finds Moonshot's Kimi K3 leads open-weight models on offensive cyber tasks yet sits far below US frontier systems.
Alibaba ships Qwen-Image-3.0 and chart-topping TTS models
Qwen-Image-3.0 renders dense infographics and ten-pixel text in one pass, while Qwen-Audio-3.0-TTS tops a speech leaderboard across 16 languages.
Cisco open-sources tiny Antares models to locate code flaws
Cisco's Foundation AI arm releases 350M and 1B open-weight models that trace known vulnerabilities to specific files, nearing frontier accuracy at a fraction of the cost.
Poolside's Laguna S 2.1 tops open-weight agentic coding class
Poolside released Laguna S 2.1, a 118B MoE coding model with 1M context under an open license, hitting 78.5% on SWE-Bench Multilingual with only 8B active params.
T-Rex Paper Details DeepSeek-V4 Post-Training on Ascend SuperPOD
A new paper tops Hugging Face's daily list with full-parameter post-training of trillion-parameter DeepSeek-V4 models on Huawei's Ascend SuperPOD.
Google ships Gemini 3.6 Flash, Flash-Lite, and a cyber model
Google's new Flash tier targets agentic workloads with fewer output tokens and lower prices, plus a cybersecurity model piloted with governments.
NVIDIA ships Cosmos 3 Edge, a 4B world model for on-device robots
NVIDIA open-sourced Cosmos 3 Edge, a 4B-parameter world model that reasons and generates robot actions locally, hitting 15 Hz real-time control on Jetson Thor.
Moonshot pauses Kimi K3 signups as demand outstrips GPUs
Moonshot AI halted new Kimi K3 subscriptions after demand surged sixfold, as third-party tests put the open-weight model within reach of Claude Fable 5.