Search “Multimodal”

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

Mistral releases Large 4 weights on Hugging Face

On October 29, 2026, Mistral released the open weights of Mistral Large 4 (Le Chonk), a 1-trillion-parameter native multimodal model that continues reinforcement learning on the basis of the preview version, with improved coding and defensive cybersecurity capabilities. According to data published by Mistral AI, the model scores 19.7% on FrontierCode 2.0 and 48% on the Artificial Analysis Coding Agent Index, roughly on par with GLM-5.3, but uses an average of 47% fewer tokens per task; on Terminal-Bench 4.0 it improves by 13 percentage points to 41%. The model supports three reasoning modes, uses a modified MIT Mistral Ecosystem Open-Weight license, and is priced at $0.70 per million input tokens, $0.07 for cached input, and $2.10 for output via API. On the same day, Unsloth released a 1-bit GGUF quantized version of the model, stating that it can run locally on a Mac Studio, retains 91% top-1 accuracy after quantization, and provides a download link and usage guide.

Diffusion Intelligence releases diffusion language model Clover-35B-A3B-Preview

On October 29, 2026, Diffusion Intelligence released the diffusion language model Clover-35B-A3B-Preview. According to QbitAI, the model uses a MoE architecture with 35 billion total parameters and about 3 billion activated per token, and supports a thinking mode; in the 49 tasks of the web Agent benchmark WindTunnel, its median execution time was 1.52 seconds, lower than the 3.21 seconds of the Jev+Mercury 2.5 combination, with success rates of 98.6% and 95.9% respectively; on code benchmarks it scored 73.6 on SWE-bench Verified and 51.7 on Terminal-Bench 2.1. The team said it will subsequently launch a multimodal version and advance on-device adaptation. On October 30, IT Home · Artificial Intelligence reported that the artificial intelligence lab DiffuSpace released Clover-35B-A3B-Preview, the first model in the Clover series, which uses a MoE architecture and is open-sourced under the Apache license, generating results through multiple rounds of iteration; DiffuSpace said it outperforms the same-scale autoregressive model Qwen3.6-35B-A3B on multiple benchmarks while maintaining faster generation speed, and is especially good at agentic coding and tool-use tasks.

Ant Group's GuangJian team releases visual forensics model DefakerOne2

Ant Group's GuangJian team has open-sourced the visual forensics model DefakerOne2 on GitHub and released the model weights. The model is built on the Qwen3.5-2B multimodal backbone with 2B parameters, covering AI-generated content, DeepFake faces, document tampering, natural image tampering, face and ID anti-counterfeiting, and AI-generated video detection. The team says the model was trained on about 50 million samples, using a three-stage Tide-Style Tuning process from LoRA to full parameters and back to LoRA, and employs entropy-based test-time routing to choose between the video and frame pathways. The team also says it achieved SOTA on 40 of 53 detection benchmarks and on 11 of 14 localization benchmarks, and that the weights and the RRVBench benchmark have been released on ModelScope.

Kimi K3.1 Pro ranks 8th on WebDev Arena

Moonshot AI releases Kimi K3.1 Pro, which ranks 8th on Code Arena WebDev with a score of 1692. According to Arena, its score ties with Claude Opus-5 (Max) at 70% lower cost, and it rises 9 places from Kimi K3 (Max)'s 17th place. The model is priced at $2/$10 per million input/output tokens, with a median cost per task of $1.23, 29% lower than Kimi K3, and its weights will be opened on November 4. Kimi officially calls it a 3T-class open-weight model, featuring a 1M token context and a 2.8T total parameter/104B activated parameter MoE multimodal architecture, with hallucination rate reduced by 65%. Testing Catalog reports that Kimi thus returns to the top 4 in the Code Arena WebDev lab rankings, and says that after the open-weight release it will become the top-ranked open-source model on that leaderboard.

fal launches the Kling 4.0 video model

fal launched Kling 4.0 on October 21, 2026. According to information released by fal, this is a new-generation native multimodal video model that improves the stability of complex motion, reference consistency for characters and scenes, and control precision for cinematic storytelling. Currently, the model is available only through fal UX and fal Agent, with API access coming soon. Apart from this launch fact, there are no other subsequent developments with qualified factual evidence.

NVIDIA Expands Nemotron Retriever Series and Releases Multiple Open Models

On October 21, 2026, NVIDIA expanded the Nemotron Retriever series, releasing multiple open models covering embedding, reranking, and document parsing. The models introduced include Nemotron 3.5 Embed (8B BF16, 1B BF16, 1B NVFP4), Nemotron 3.5 Rerank 8B BF16, and Nemotron 3.5 Parse with 631 million parameters, aimed at multimodal retrieval and document processing for enterprise agent workflows. NVIDIA says that adding Rerank to the agent retrieval pipeline reduces the token cost per query by about 14% (with the 8B embedding model) and 13% (with the 1B model), respectively, while also improving retrieval accuracy. Parse scores 75.20 on ParseBench and processes 10.5 pages per second on a single H100. Open weights, data, and training recipes have been released on Hugging Face and can be deployed via NVIDIA NIM and through Baseten, DeepInfra, and FriendliAI.

Perplexity Decider launches on Vercel AI Gateway

On October 17, 2026, Perplexity's Decider V1.2 decision model launched on Vercel AI Gateway. Vercel developers can now use perplexity/pplx-decider-v1.2 for complex agentic workflows and perplexity/pplx-decider-flash-v1.2 for ultra-fast decisions. Aravind Srinivas reposted the news. Earlier the same day, reports said Perplexity had open-sourced two decision models, pplx-decider-flash-v1.2 and pplx-decider-v1.2: the former has 4B parameters and costs 1 cent per million tokens, and is called the strongest 4B model on the HF Decision Index; the latter has 27B parameters and costs 2 cents per million tokens, and ranks 2nd on the overall HF Decision Index leaderboard. Both are now available on the Perplexity Decisions API and offer open weights on Hugging Face. Perplexity Developers said this generation of open-weight multimodal decision models improves both text and vision scores, delivers inference up to 5 times faster, and completes all questions in a single request in one forward pass.

Meta releases open-weight model Muse Spark Open

On October 15, 2026, Meta released the open-weight model Muse Spark Open, with the weights going live on Hugging Face the same day under the Muse Model License. According to the PyTorch Blog, the model has 816B total parameters, 42B activated parameters, and a 1M token context window; it is natively multimodal and aimed at coding and agentic tasks, adds safety post-training on top of Muse Spark, and can run on SGLang and vLLM. On October 16, the model launched on Amazon SageMaker JumpStart with two variants, BF16 and NVFP4; the same day it entered live testing in Arena's Agent Arena and launched in Code Arena's WebDev and other projects. On October 17, Arena announced that Muse Spark Open (Max) ranked first among US open models on the Code Arena WebDev leaderboard with a score of 1511, placing 13th among all open models.