IT Home · Artificial IntelligenceModel releases
AI score75MiniMax Open-Sources M3.1 Flash Model Weights with 428B Total Parameters
English text is being prepared.
#Model releases#Open-source ecosystem#English text is being prepared.
NAll news
All news
- IT Home · Artificial IntelligenceModel releasesAI score71
ByteDance's Volcano Engine releases the Seedance 2.5 4K model, targeting film-grade image generation
Volcano Engine, under ByteDance, has released Seedance 2.5 4K, which through specialized training learns character expressions, subtle movements, and image details, making it the most powerful model in the Seedance 2.5 generation. The model supports a Draft sample mode, allowing users to first validate storyboards and refine shots with 480P samples before generating the final 4K footage, and it features 10-bit color depth. It is currently available only for trial by select enterprise customers, and some domestic and overseas film and television production and post-production companies have already tried it.
#Model releases#Video#Multimodal
- cnBetaModel releasesAI score69
Hugging Face Launches Carbon-X, an Open-Source DNA Model Based on Qwen3-14B
Hugging Face, together with research labs in Italy and China, launched the open-source model Carbon-X, which is trained on genomic sequences and natural language based on Qwen3-14B and can answer questions about DNA sequences, predict the effects of mutations, and generate new sequences. Hugging Face says that on five DNA understanding and generation benchmarks, Carbon-X's average score is close to GPT-6.1 Sol, while its inference cost is only 1/1800 of the latter, and it scores 27 points higher than Qwen3-14B. The model can run locally, and its training data explicitly excludes virus-related data.
#Model releases#Open-source ecosystem#English text is being prepared.
- MiniMaxFirst-partyModel releasesAI score79
MiniMax M3.1 Flash releases open-source weights, 428B total parameters, 23B activated per token
MiniMax M3.1 Flash: Space Performance + Bunny Speed 🚀🐰 Yes, Space Bunny - which has topped global leaderboards with 200T+ tokens - was an earlier version of M3.1 Flash. 💰Redefining the performance economics of agentic AI. M3.1 Flash achieves frontier-level intelligence across key tasks such as Cowork (Automation-Bench: 62%; ALE: 30%), Coding (DeepSWE: 74%) and Visual Reasoning (BabyVision: 91%), at 1/3 the cost of M3 and 3x the speed. API pricing (flat rate): $0.15 / $0.60 per 1M input/output tokens, with cache reads at $0.015 / 1M tokens. Model weights will arrive on Oct 23. API: https://t.co/fHRdSV73Hr M Plan: https://t.co/XfJWSyauda MiniMax Code for desktop: https://t.co/SGmLTI2HQm
#Model releases#Open-source ecosystem#Agents
- Viggle AIModel releasesAI score58
Viggle AI releases world foundation model JST-2
#Model releases#Video#Multimodal
- Unite.AIModel releasesAI score79
Canva Releases Its In-House Image Model Canva Craft
On October 22, Canva released Canva Craft, the first foundation model trained from scratch by its internal lab Canva Research. It is now available to Canva users worldwide through Dream Lab and to developers via a public API. The model uses a two-stage architecture: first a planning model parses the prompt and generates a composition plan, then a rendering model outputs images at up to 2K resolution supporting more than 30 aspect ratios, and can generate text in 55 languages. Canva says Craft debuted fourth on DesignArena, and previews that two faster variants, Craft-fast and Craft-ultrafast, are coming soon.
#Model releases#Image generation#English text is being prepared.
- CNETModel releasesAI score66
Hugging Face releases the Carbon-X model, enabling conversation with DNA sequences in natural language
Hugging Face has released Carbon-X, the newest and most capable model in the Carbon series; researchers can input DNA sequences and ask questions in natural language to obtain interpretations, predictions, or explanations. Researcher Lewis Tunstall said the model is best at predicting the effects of mutations and completing DNA sequences, for example determining whether a given sequence change might be harmful, and can run sensitive or unpublished sequences on local hardware. Hugging Face also noted limitations such as counting errors, fabricating sequence identity or function when prompt information is insufficient, and degraded output quality on long sequences, and it publicly released the training recipe for community scrutiny.
#Model releases#Open-source ecosystem#English text is being prepared.
- Phoenix TechModel releasesAI score71
Hugging Face launches Carbon-X, a 14-billion-parameter DNA model built on Qwen3-14B
Hugging Face announced on Thursday the launch of Carbon-X, an open-source model with 14 billion parameters for DNA understanding and generation, obtained by training Qwen3-14B on genomic sequences and natural language. It integrates DNA sequences and natural language into a single model, allowing researchers to ask questions about specific DNA sequences, predict the effects of mutations, and generate new DNA sequences; it can also serve as a sub-agent to help other models interpret DNA sequences. Hugging Face says that on five DNA understanding and generation benchmarks, Carbon-X outperforms all other open-source models it evaluated, with an average score close to GPT-6.1 Sol, but at an inference cost per question of only 1/1800 of the latter, and 27 points higher than Qwen3-14B.
#Model releases#Open-source ecosystem#English text is being prepared.
- Artificial AnalysisModel releasesAI score62
Ant Group's Ming-3.0-tts achieves top-tier speech synthesis quality with 0.5B parameters
#Evaluations / benchmarks#Speech#Open-source ecosystem
- Robotics & Automation NewsModel releasesAI score82
Google DeepMind releases Gemini 4 Argon, which can teach the Apollo humanoid robot new skills from video
Google DeepMind releases Gemini 4 Argon and demonstrates the model learning physical tasks by watching videos, then transferring them to Apptronik's Apollo humanoid robot. In internal tests, after watching a video of beanbag throwing, Argon analyzed the environment, mapped spatial coordinates, and computed full-body joint motions, autonomously generating simulation code; Apollo hit the target seven times in a row in real trials. Google DeepMind says this Video-to-Sim-to-Real pipeline can compress simulation modeling that originally took weeks into hours, Argon scored 87% on the embodied reasoning benchmark ERQA, and the model has been opened through Google's AI products and the Gemini API.
#Model releases#Embodied AI#Multimodal
- ThurrottModel releasesAI score69
Canva Releases Its First In-House Image Model, Craft
Canva has released its first in-house image model, Craft, trained from scratch by Canva Research and built specifically for design rather than photography. It can generate 2K images in any aspect ratio and supports editable output. Canva says it ranks fourth on DesignArena, behind only three OpenAI models: GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare, and GPT Image 2. Craft is now available to paid subscribers within the Canva app and to developers through a public API, with Canva MCP Server support coming soon; two variants, Craft-fast and Craft-ultrafast, will also be released later.
#Model releases#Image generation
- falModel releasesAI score75
fal Releases FLUX 3 Max Video Model, Generating a 10-Second Video Takes 12.45 Seconds
Introducing FLUX 3 Max, the latest post-trained video model from fal Research in partnership with @bfl_ai. FLUX 3 Max is tied for #1 in image-to-video quality on independent third-party benchmarks, generating a 10-second, 720/768p video in ~12 seconds. For the next three weeks, FLUX 3 Max is 75% off, making it the most affordable frontier video model at its quality level.
#Model releases#Video#Reasoning
- Rohan PaulModel releasesAI score74
Canva releases its first in-house image model, Canva Craft
Introducing Canva Craft 🚀 Extraordinary image generation is now available to everyone, powered by our first frontier model built specifically for design. With Craft, Canva is now the world's second most advanced AI lab for image generation on @DesignArena's global leaderboard. See what it can do 👇
#Model releases#Image generation#OpenAI
- Thomas WolfModel releasesAI score69
Hugging Face, together with TIGEM and other institutions, releases the open-source DNA model Carbon-X
#Model releases#Open-source ecosystem#English text is being prepared.
- TechCrunch AIModel releasesAI score65
Canva launches its first image generation model, Canva Craft
Canva has launched its first image generation model, Canva Craft, developed by its in-house research team and benchmarked against similar models from Google, OpenAI, ByteDance and Meta. The model works in two stages: it first generates a plan containing spatial relationships, relative proportions, visual hierarchy and text information, then outputs images at up to 2K resolution supporting about 30 aspect ratios, with an emphasis on accurate object details, clear text and precise local editing. The model is trained on human-designed works from Canva's template library rather than photos, and the generated results can be converted into layered designs via Canva Magic. It is now available to all users on the Dream Lab tab and will be integrated into third-party chatbots such as ChatGPT and Gemini through the Canva API and MCP.
#Model releases#Image generation#Multimodal
- Zhi DongxiModel releasesAI score66
ElevenLabs Releases Eleven Realtime Real-Time Voice Model, Hands-On Test of Voice-Built Agent
On October 20, ElevenLabs released the Eleven Realtime series of real-time voice models, including v1 Mini for natural conversation and v1 Max for customer service and multi-step tasks, supporting more than 90 languages and 17,000 voices. According to Artificial Analysis testing, Max achieved 98.8% accuracy on the Big Bench Audio voice reasoning benchmark, ranking third, with an average first audio response time of 0.57 seconds, while Mini was 0.62 seconds. Both models can be used through the ElevenAgents platform or the Realtime API compatible with the OpenAI Realtime specification, and currently only support WebSocket connections; Zhi Dongxi's hands-on test found that its Architect building assistant can modify Agent settings by voice, but the Chinese accent is noticeable, responses tend to be long, and latency is relatively high.
#Model releases#Speech#Agents
- ZhidxModel releasesAI score78
Recursive, co-founded by Yuandong Tian, open-sources Flint, a 40B local coding model
Recursive Superintelligence, co-founded by former Meta FAIR research director Yuandong Tian, has launched its first local coding model, Flint, with 39.8 billion total parameters and approximately 2.5 billion activated parameters per token. It has been open-sourced under the Apache 2.0 license, with weights available via Hugging Face. The model generates over 150 tokens per second on NVIDIA RTX Spark and over 125 tokens per second on a MacBook Pro with M4 Pro, and scores 54.9%, 89.7%, and 74.2% on Terminal-Bench 2.1, SWE-Bench Pro v2, and SWE-Bench Verified, respectively. Recursive says its team completed the entire training pipeline from scratch in under 4 months with the help of its automated research system Eureka Machine.
#Model releases#Open-source ecosystem#English text is being prepared.
- ZhidxModel releasesAI score71
iFlytek releases the Spark X2.6 cloud-device integrated large model, trained on fully domestic computing power
iFlytek releases the cloud-device integrated large model iFlytek Spark X2.6, including the cloud-side 293-A30B model and the already open-sourced on-device X2.6-4B model, both of which completed capability upgrades based on fully domestic computing power. The cloud model supports a million-token context, with an input price of 0.9 yuan per million tokens and an output price of 3.6 yuan per million tokens; its input cache-miss and output prices are lower than those of DeepSeek-V4.1-Flash-1017. At the same event, it also released the speech foundation model SparkAudio 1.0, the Robot Super Brain Platform V3.0, the WinWork digital employee platform, and hardware such as the iFlytek AI recording ring.
#Model releases#English text is being prepared.#Speech
- MarkTechPostModel releasesAI score68
Gradium Open-Sources Phonon On-Device TTS Models, 31M Version Uses Apache 2.0
Gradium has open-sourced the Phonon series of on-device text-to-speech models. The English version, Phonon 31m, has 31M parameters and a 64.5 MB q8 GGUF file, uses the Apache 2.0 license, and outputs 24 kHz mono audio. The model runs without a GPU on CPUs, phones, Raspberry Pi 5, and in browsers. On a MacBook Air M5 CPU, it produces the first audio segment in 6.9 ms, and on Gradium's self-tested 500-sentence hard set it achieves a WER of 1.70%, compared with 3.50% for Kokoro. The 40m version covers five languages—English, French, German, Spanish, and Portuguese—but uses the CC BY-NC 4.0 non-commercial license.
#Model releases#Open-source ecosystem#Speech
- Testing CatalogModel releasesAI score69
Ideogram Releases Vector Model Vector 1.0, Capable of Generating Editable SVGs and Vectorizing Images
Today, we’re introducing Ideogram's first vector model. Generate clean, editable SVGs with smooth gradients and sharp typography from a text prompt, or turn any image into a detailed, scalable vector. Live in Ideogram and the API: https://t.co/Cou2mo5p37 https://t.co/TWBuLiJ31a
#Model releases#Image generation#Multimodal