# New AI model releases

Announced models isitsota tracks that are still waiting for independent benchmarks, newest first. 98 pending as of Oct 22, 2026. Lab claims are not used in any verdict.

RSS: https://isitsota.com/releases/rss.xml

- [Carbon-X](https://isitsota.com/m/carbon-x) (Hugging Face, announced Oct 22, 2026) — An open model from Hugging Face that takes raw DNA and natural language in one chat: it describes sequences, predicts whether a mutation is benign or…
- [Chariot-1-ASR Realtime](https://isitsota.com/m/chariot-1-asr-realtime) (Chariot, announced Oct 22, 2026) — Chariot's streaming speech-recognition model for Indian languages
- [Ideogram Vector](https://isitsota.com/m/ideogram-vector) (Ideogram, announced Oct 22, 2026) — Ideogram's first vector model
- [Ling-3.0-flash-Sante](https://isitsota.com/m/ling-3-0-flash-sante) (Ant Ling (InclusionAI, Ant Group), announced Oct 22, 2026) — Ant Ling's healthcare model from Ant Group's InclusionAI, built on Ling-3.0-flash: a mixture-of-experts model with 124B total and 5.1B active parameters…
- [Ling-3.1-flash-1019](https://isitsota.com/m/ling-3-1-flash-1019) (Ant Ling (InclusionAI, Ant Group), announced Oct 22, 2026) — A new checkpoint of Ant Ling's Ling-3.1-flash from Ant Group's InclusionAI, aimed at agentic work
- [P-Video-2-Pro-Avatar](https://isitsota.com/m/p-video-2-pro-avatar) (Pruna AI, announced Oct 22, 2026) — Pruna AI's talking-avatar model, built on MiniMax H3
- [PixAI Tagger 1.0](https://isitsota.com/m/pixai-tagger-1-0) (PixAI, announced Oct 22, 2026) — PixAI's open anime image tagger, the model PixAI used to curate training data for its Tsubaki.3 image model
- [PixText](https://isitsota.com/m/pixtext) (ATH-MaaS, announced Oct 22, 2026) — A model that fixes small text inside generated images directly in pixel space: give it an image, the regions to edit and the target text, and it redraws only…
- [Clara](https://isitsota.com/m/clara-saanora) (Saanora, announced Oct 21, 2026) — Saanora's decision model
- [Cohere Rerank 5](https://isitsota.com/m/cohere-rerank-5) (Cohere, announced Oct 21, 2026) — Cohere's fifth-generation reranking models for search, RAG and agents, in Pro (highest quality) and Fast (low latency, high throughput) tiers
- [Numo-1-E2B](https://isitsota.com/m/numo-1-e2b) (Mycel, announced Oct 21, 2026) — Mycel's small open model for decisions about scientific papers
- [NVIDIA Kumo](https://isitsota.com/m/nvidia-kumo) (NVIDIA, announced Oct 21, 2026) — NVIDIA's family of open foundation models for structured business data, presented at GTC Berlin as Kumo Structured, part of the NVIDIA Agent Knowledge…
- [OXR-1.0](https://isitsota.com/m/oxr-1-0-somark) (SoMark, announced Oct 21, 2026) — SoMark's open-weight document parsing model
- [Pangaea](https://isitsota.com/m/pangaea-emet) (Emet by Deel, announced Oct 21, 2026) — Deel's family of language models for payroll, HR and compliance questions, trained with data and tests from Emet, Deel's new AI lab for model evaluation and…
- [PatentSec-12B](https://isitsota.com/m/patentsec-12b) (Calera Labs, announced Oct 21, 2026) — Calera Labs' patent-prosecution model, a fine-tune of Google's Gemma 4 12B instruction model for USPTO statutory reasoning: claim-element auditing…
- [Raon-1 mini (preview)](https://isitsota.com/m/raon-1-mini-krafton) (KRAFTON AI, announced Oct 21, 2026) — KRAFTON AI's first general language model, a sparse mixture-of-experts model trained from scratch, released as a preview
- [Recursive Flint](https://isitsota.com/m/flint-recursive) (Recursive, announced Oct 21, 2026) — Recursive's first model: a mixture-of-experts coding model for local use, built with the company's automated research system (the Eureka Machine)
- [Sage-A1.5](https://isitsota.com/m/sage-a1-5) (InternScience (Shanghai AI Laboratory), announced Oct 21, 2026) — An open agentic model for long-horizon scientific work from InternScience at Shanghai AI Laboratory, fine-tuned from Qwen3.6-35B-A3B
- [Simba 3.5](https://isitsota.com/m/simba-3-5-speechify) (Speechify, announced Oct 21, 2026) — Speechify's text-to-speech model that speaks 48 languages under a single model id, with every catalog voice and cloned voices usable in any of them
- [Topview T1](https://isitsota.com/m/topview-t1) (Topview, announced Oct 21, 2026) — Topview's video generation model, available on the Topview platform for text-to-video, image-to-video and multi-reference generation at 720p or 1080p, in clips…
- [CAPHARD](https://isitsota.com/m/caphard-phagos) (Phagos, announced Oct 20, 2026) — Phagos' phage–bacteria interaction model
- [Spark-X2.6-4B](https://isitsota.com/m/spark-x2-6-4b) (iFlytek, announced Oct 20, 2026) — iFlytek SparkLLM's compact open model for on-device agents
- [Underdog Bark 3.0](https://isitsota.com/m/bark-3-underdog) (Underdog, announced Oct 20, 2026) — Underdog's small open English speech-to-text model, built to run on Macs and phones
- [Ampersand](https://isitsota.com/m/ampersand-humans-and) (humans&, announced Oct 20, 2026) — humans&'s collaborator model, trained to help people and groups reach their individual and collective goals over long, open-ended interactions
- [ApplePie](https://isitsota.com/m/applepie-general-data) (The General Data Corporation, announced Oct 20, 2026) — General Data's open PII detection model built for Europe: it tags 20 kinds of personal data, including national ID formats (French NIR, German Steuer-ID…
- [BioEmu-2](https://isitsota.com/m/bioemu-2) (Microsoft, announced Oct 20, 2026) — Microsoft Research AI for Science's foundation model for biomolecular structure, dynamics and function, in public preview
- [Cara-4.1 Express](https://isitsota.com/m/cara-4-1-express) (Anam, announced Oct 20, 2026) — Anam's real-time avatar model, built on Cara-4 to be faster and more expressive, with better lip sync, Director Notes style prompts and wider framing
- [Cosmos3-H-Embodied-Nano](https://isitsota.com/m/cosmos3-h-embodied-nano) (NVIDIA, announced Oct 20, 2026) — NVIDIA's 16B surgical-robotics model for dVRK-style and other surgical robots
- [Deployxa LM 1.0](https://isitsota.com/m/deployxa-lm-1-0) (Deployxa, announced Oct 20, 2026) — Deployxa's 1.5B open-weights SRE model that diagnoses container and deployment failures and returns machine-executable JSON fix recipes
- [Eleven Realtime](https://isitsota.com/m/eleven-realtime) (ElevenLabs, announced Oct 20, 2026) — ElevenLabs' new class of speech-to-speech models for voice agents, which the lab says starts talking in about half a second
- [FastH3 Omni](https://isitsota.com/m/fasth3-omni) (Hao AI Lab (UC San Diego), announced Oct 20, 2026) — Hao AI Lab's FastVideo team's 8-step reference-to-video model, distilled from MiniMax's MiniMax-H3: video with synchronized audio from a text prompt plus…
- [Fibo 2.0](https://isitsota.com/m/fibo-2-0) (Bria AI, announced Oct 20, 2026) — Bria AI's image generation and editing release, trained only on licensed data (lab statement)
- [Flora](https://isitsota.com/m/flora-nvidia) (NVIDIA, announced Oct 20, 2026) — NVIDIA GEAR lab's suite of open models that watch robot and human video and turn it into training signals: timestamped captions, dense progress rewards and…
- [Maximo Atlas 1.5](https://isitsota.com/m/maximo-atlas-1-5) (Maximo AI, announced Oct 20, 2026) — Maximo AI's first sparse model, aimed at agentic coding, UI design and real-time 3D from a single prompt
- [MemT-1.0](https://isitsota.com/m/memt-1-0) (MemTensor, announced Oct 20, 2026) — MemTensor's open-weight model family (Flash and Nano) with persistent memory built into model computation, so agents can reuse past history without reloading…
- [Monarch 2](https://isitsota.com/m/monarch-2) (Africa Compute Fund, announced Oct 20, 2026) — Africa Compute Fund's general-purpose LLM with an emphasis on economic, institutional and development knowledge, including African economies, languages and…
- [Naad v2](https://isitsota.com/m/naad-v2) (Oogam AI, announced Oct 20, 2026) — Oogam AI's speech-to-text and text-to-speech models for 28 Indian languages (3 in beta), offered through the Oogam API
- [NOVA (Zenithon)](https://isitsota.com/m/nova-zenithon) (Zenithon, announced Oct 20, 2026) — Zenithon calls NOVA the first world model for extreme physics: a single transformer pretrained on plasma and fluid physics systems that predicts how complex…
- [Omnii-1](https://isitsota.com/m/omnii-1) (Radical Numerics, announced Oct 20, 2026) — Radical Numerics' genome language model, pretrained across 28 biological modalities including DNA, RNA, proteins and epigenomics, and adapted to eight…
- [Ornith-1.5-122B](https://isitsota.com/m/ornith-1-5-122b) (Ornith, announced Oct 20, 2026) — Ornith's 122B mixture-of-experts model for coding and agentic work, with open weights under the MIT license
- [Phonon](https://isitsota.com/m/phonon-gradium) (Gradium, announced Oct 20, 2026) — Gradium's suite of on-device text-to-speech models under 50M parameters, built to run offline on phones, laptops, small boards and in the browser
- [Pocket TTS 4.0](https://isitsota.com/m/pocket-tts-4-0) (Kyutai, announced Oct 20, 2026) — Kyutai's Pocket TTS v4.0.0 release: six small English text-to-speech models with voice cloning, at 18.2M, 6.9M and 1.9M parameters, each with a…
- [RFD4-Proteína](https://isitsota.com/m/rfd4-proteina) (NVIDIA, announced Oct 20, 2026) — NVIDIA and the Baker Lab's flow-matching model for biomolecular design
- [Sentiment-Analysis-7B](https://isitsota.com/m/microsoft-sentiment-analysis-7b) (Microsoft, announced Oct 20, 2026) — Microsoft's 7B open-weights model, fine-tuned from Mistral-7B-Instruct-v0.2, that labels a user message about a service, product or agent as Negative, Neutral…
- [VideoGen Model 1](https://isitsota.com/m/videogen-model-1) (VideoGen, announced Oct 20, 2026) — VideoGen's video model that turns a brief, with optional reference images, into a finished multi-shot video with audio, up to ten minutes long (lab statement)
- [Voxtral 3](https://isitsota.com/m/voxtral-3) (Mistral, announced Oct 20, 2026) — Mistral's new speech-to-text and text-to-speech models, with a voice agent SDK built on LiveKit and Pro Voice cloning from up to 30 minutes of audio
- [SenseNova-RoboRSI-27B-Preview](https://isitsota.com/m/sensenova-roborsi-27b-preview) (SenseTime, announced Oct 19, 2026) — SenseTime's 27B embodied model for robot planning and decision-making, trained from Qwen3.8-27B with experience from its open-source SenseNova-RoboRSI harness…
- [Atomwell](https://isitsota.com/m/atomwell) (Bindwell, announced Oct 19, 2026) — Bindwell's foundation model for predicting experimental outcomes in molecular discovery, trained across 800M–9B sizes on about 80M experimental measurements
- [Clairic](https://isitsota.com/m/clairic) (Cloudflare, announced Oct 19, 2026) — Cloudflare's System One extraction model, which extracts, computes, estimates and forecasts values from unstructured inputs
- [Dream-1](https://isitsota.com/m/dream-1) (Dream Labs, announced Oct 19, 2026) — Dream Labs' World Action Model, a robot foundation model that the lab says controls unseen bimanual parallel-gripper robots zero-shot on unseen tasks
- [Drex Max](https://isitsota.com/m/drex-max) (Nace.AI, announced Oct 19, 2026) — Nace AI's multimodal decision model for text, images and video, available through Nace at $0.06 per 1M input tokens
- [FLUX 3 Max](https://isitsota.com/m/flux-3-max) (fal, announced Oct 19, 2026) — A post-trained image-to-video model from fal Research, built in partnership with Black Forest Labs and available on fal
- [GoTo-AI-Gemma-3-27B-IT](https://isitsota.com/m/goto-ai-gemma-3-27b-it) (GoTo (with AI Singapore), announced Oct 19, 2026) — GoTo's 27B Gemma 3 instruct model for English and Indonesian regional languages (Javanese, Sundanese, Batak Toba, Balinese), developed with AI Singapore
- [Mercury 3](https://isitsota.com/m/mercury-3) (Inception, announced Oct 19, 2026) — Inception's diffusion LLM with a 1M-token context, live on the Inception API, OpenRouter, Baseten and Vercel
- [Ming-3.0](https://isitsota.com/m/ming-3-0) (Ant Ling (InclusionAI, Ant Group), announced Oct 19, 2026) — InclusionAI's open Ming-3.0 audio family: multilingual TTS with voice cloning, Chinese/English ASR with dialects, and native audio-language MoE models for…
- [MiST](https://isitsota.com/m/mist) (Dream, announced Oct 19, 2026) — Dream's open cybersecurity language models (8B and 30B-A3B), with a dedicated security mid-training stage before chat tuning
- [MJev-Spatial](https://isitsota.com/m/mjev-spatial) (Mind Lab (Macaron AI), announced Oct 19, 2026) — Mind Lab's System 1 spatial decision model on Qwen3.5-4B: scores answer options, value bins and image regions with native attention Q/K, without autoregressive…
- [Morpheus-2.0](https://isitsota.com/m/morpheus-2-0) (JPMorgan Chase, announced Oct 19, 2026) — A time-series forecasting foundation model from a JPMorgan Chase team, announced in a thread by a team member
- [Ropedia W0](https://isitsota.com/m/ropedia-w0) (Ropedia, announced Oct 19, 2026) — Ropedia's egocentric world foundation model trained on about 10k hours of first-person video for loco-manipulation that transfers across robot embodiments
- [SKY-001](https://isitsota.com/m/sky-001) (Skyfall AI, announced Oct 19, 2026) — Skyfall AI's Engineering World Model for CAD and BIM tasks
- [Smaug-AI-Image-Detector](https://isitsota.com/m/smaug-ai-image-detector) (Abacus.AI, announced Oct 19, 2026) — Abacus.AI's open-weights detector for AI-generated images, released on Hugging Face under Apache 2.0 with GGUF builds
- [Viggle JST-2](https://isitsota.com/m/viggle-jst-2) (Viggle AI, announced Oct 19, 2026) — Viggle's physics-native 3D world foundation model: scenes and characters with motion, built in 3D rather than pixel/video space
- [EdenLive](https://isitsota.com/m/edenlive) (VAST, announced Oct 18, 2026) — VAST's real-time world-model renderer that turns a rough game or reference video into detailed frames while upstream code keeps the world state
- [Agens Volundr 32B v1](https://isitsota.com/m/agens-volundr-32b-v1) (Blockway, announced Oct 18, 2026) — Blockway's open-weights 32B model for agentic coding, the full release after its Preview, tuned for long agent sessions
- [OrcaCyber Nano 1.0](https://isitsota.com/m/orcacyber-nano-1-0) (OrcaRouter, announced Oct 18, 2026) — A 27B open-weight vision-capable model built for defensive cybersecurity work such as vulnerability research and incident response
- [Yug-1](https://isitsota.com/m/yug-1) (Birla AI Labs (Aditya Birla Group), announced Oct 17, 2026) — Birla AI Labs' first time-series foundation model: a 271.8M-parameter model for zero-shot probabilistic forecasting that returns a range of likely values…
- [REKI](https://isitsota.com/m/reki) (Kyutai, announced Oct 16, 2026) — Kyutai's 2.4B vision-language-action robot model, built with Inria and École des Ponts, pretrained on public data only, with a built-in memory of its recent…
- [NVIDIA Nemotron Labs Dory 1.0 4B](https://isitsota.com/m/nvidia-nemotron-labs-dory-1-0-4b) (NVIDIA, announced Oct 16, 2026) — NVIDIA's 4B looped model: the same weights can run 1, 2 or 4 loops per token, trading compute for accuracy
- [GLiClass v4](https://isitsota.com/m/gliclass-v4) (Knowledgator, announced Oct 16, 2026) — Knowledgator's family of instruction-following classifier encoders, from a 435M Edge model up to a 27B model
- [Glimpse](https://isitsota.com/m/glimpse) (Elorian AI, announced Oct 16, 2026) — Elorian AI's first frontier visual-reasoning model, built for visual thinking
- [Hy-Memory-30B-A3B](https://isitsota.com/m/hy-memory-30b-a3b) (Tencent, announced Oct 16, 2026) — Tencent Hunyuan's dedicated long-term memory writer for AI agents, part of Hy Memory 2.0 alongside a framework and MemHub
- [LemonSlice Cartoon Model](https://isitsota.com/m/lemonslice-cartoon-model) (LemonSlice, announced Oct 16, 2026) — Animates expressive, interactive cartoon avatars from a single image
- [MOSS-Image](https://isitsota.com/m/moss-image) (OpenMOSS / MOSI, announced Oct 16, 2026) — OpenMOSS / MOSI's text-to-image Diffusion Transformer, released as MOSS-Image-Base and MOSS-Image-SFT, with Chinese and English text rendering and controllable…
- [StartLux-Decision](https://isitsota.com/m/startlux-decision) (StartLux, announced Oct 16, 2026) — StartLux's open-weight decision-model family: an Instant line from 0.8B to 27B and an AutoThink adaptive-reasoning line from 4B to 27B
- [Halcyon](https://isitsota.com/m/halcyon-radical-numerics) (Radical Numerics, announced Oct 15, 2026) — Radical Numerics' family of early-fusion autoregressive text-to-image base models, trained entirely from scratch on 100M image-text pairs (GPIC) with no…
- [Matrix-Game 4.0](https://isitsota.com/m/matrix-game-4-0) (Skywork AI, announced Oct 15, 2026) — Skywork's real-time video world model for games: a game engine keeps the world state and rules, and the video model renders each frame
- [NABLA-FC](https://isitsota.com/m/nabla-fc) (NABLAS Inc., announced Oct 15, 2026) — NABLAS's Japanese fact-checking model (Qwen3-8B fine-tune) plus a citing agent that surfaces primary sources with the verdict; UTokyo spin-off work from GENIAC…
- [Chandra 2.5](https://isitsota.com/m/chandra-2-5) (Datalab, announced Oct 15, 2026) — Datalab's new OCR model, live on its API
- [Granite Switch 4.1](https://isitsota.com/m/granite-switch-4-1) (IBM, announced Oct 15, 2026) — IBM's 8B and 30B Granite Switch models: one checkpoint that switches through control tokens between chat, a general Decision adapter, a Guardian safety adapter…
- [Jais 3](https://isitsota.com/m/jais-3) (IFM (MBZUAI) with Inception42, announced Oct 15, 2026) — A family of six open, Arabic-native models for agentic work, 0.9B to 375B, built on IFM's K2 Horizon
- [Timestral](https://isitsota.com/m/timestral) (Mistral, announced Oct 15, 2026) — Mistral's first family of time-series forecasting models: zero-shot foundation models in 5M, 28M, 123M and 209M sizes, with open weights on Hugging Face
- [TYGER](https://isitsota.com/m/tyger) (NineNineSix, announced Oct 15, 2026) — NineNineSix's open full-duplex voice model, with the TYGER S2S agent
- [Bonsai 2](https://isitsota.com/m/bonsai-2) (PrismML, announced Oct 15, 2026) — PrismML's ternary on-device models in 2B, 4B and 9B sizes, based on Qwen3.5, with reasoning, vision and tool calling
- [FLUX 3 Image Klein](https://isitsota.com/m/flux-3-image-klein) (Black Forest Labs, announced Oct 15, 2026) — Black Forest Labs' 5.6B open-weights image model with native 4K output, bounding boxes and pixel-perfect editing
- [Kling 4.0](https://isitsota.com/m/kling-4-0) (Kling, announced Oct 15, 2026) — Kling AI's video generation model, now fully available after early access through kling.ai, MCP and a CLI
- [GPT-Live-1.1](https://isitsota.com/m/gpt-live-1-1) (OpenAI, announced Oct 15, 2026) — OpenAI's voice model update for Voice in ChatGPT, Codex and dots, with better accents and new voices in French, German, Hindi, Mexican Spanish, Korean and…
- [Nemotron 3.5 Bolt 24GB](https://isitsota.com/m/nemotron-3-5-bolt-24gb) (NVIDIA, announced Oct 15, 2026) — NVIDIA's local-agent model that fits into 24GB of memory, optimized for RTX Spark and meant to run alongside your apps
- [Nemotron 3.5 Super VL](https://isitsota.com/m/nemotron-3-5-super-vl) (NVIDIA, announced Oct 15, 2026) — NVIDIA's open multimodal model combining reasoning and tool use with text, image, document and video understanding, built for long-running AI agents
- [XGEN-JING](https://isitsota.com/m/xgen-jing) (XGEN Labs, announced Oct 14, 2026) — XGEN Labs' egocentric interactive experience model on MiniMax-H3: first-person video and audio from actions, reference images and observation history
- [Grok Voice 3.0](https://isitsota.com/m/grok-voice-3-0) (xAI, announced Oct 14, 2026) — Speech-to-speech successor to Grok Voice 2.0, available in the API and Voice Agent Builder
- [Ling-3.1-flash](https://isitsota.com/m/ling-3-1-flash) (Ant Ling (InclusionAI, Ant Group), announced Oct 14, 2026) — Ant Group's open-source mixture-of-experts model from the Ant Ling team, built to turn rough briefs into pages, mobile UIs and prototypes, with up to a…
- [Mirelo SFX 2.0](https://isitsota.com/m/mirelo-sfx-2-0) (Mirelo, announced Oct 14, 2026) — Mirelo's video-to-sound model
- [Ultramarine 1 Flash](https://isitsota.com/m/ultramarine-1-flash) (Paper Instruments, announced Oct 14, 2026) — Paper Instruments' agentic model for professional knowledge work in finance, law, consulting and operations, post-trained from GLM-5.3 Flash with large-scale…
- [CJRouter Clef Decision](https://isitsota.com/m/cjrouter-clef-decision) (CJRouter, announced Oct 13, 2026) — CJRouter's open Qwen-based decision models (9B and 27B) with Clef-style heads that return continue/escalate/stop-style choices and calibrated probabilities
- [Drama 3](https://isitsota.com/m/drama-3) (Fish Audio, announced Oct 13, 2026) — Fish Audio's open-weight text-to-speech model
- [LFM2.5-24B-A2B](https://isitsota.com/m/lfm2-5-24b-a2b) (Liquid AI, announced Oct 13, 2026) — Liquid AI's local model for laptops and PCs, built to handle multi-step agent tasks
- [Qwen-Image-2.1-Turbo](https://isitsota.com/m/qwen-image-2-1-turbo) (Alibaba / Qwen, announced Oct 9, 2026) — An accelerated checkpoint of Qwen-Image-2.1 that creates and edits images in 8 denoising steps, on the same 7B visual generation architecture
- [Beam](https://isitsota.com/m/beam) (Reflection AI, announced Oct 5, 2026) — Reflection AI's mixture-of-experts model for coding and agentic work, first announced Oct 5 and released with open weights under Apache 2.0 on Oct 16