open-source

IBM released Granite Embedding 311M Multilingual R2, a 311-million parameter dense embedding model with 32,768-token context length and support for 200+ languages. The model scores 64.0 on Multilingual MTEB Retrieval (18 tasks), an 11.8-point improvement over its predecessor, and ships with ONNX and OpenVINO models for production deployment.

IBM has released the Granite 4.1 family of language models under Apache 2.0 license. The models come in 3B, 8B, and 30B parameter sizes. Unsloth has released 21 GGUF quantized variants of the 3B model ranging from 1.2GB to 6.34GB.

IBM released Granite 4.1 30B, a 30-billion parameter instruction-following model with a 131,072 token context window. The model scores 80.16 on MMLU 5-shot and 88.41 on HumanEval pass@1, with enhanced tool-calling capabilities following OpenAI's function definition schema.

IBM has released Granite 4.1 8B, an 8-billion-parameter decoder-only language model with a 131,072-token context window. The model supports 12 languages and costs $0.05 per million input tokens and $0.10 per million output tokens, available under the Apache 2.0 license.

IBM has released Granite 4.1-8B, an 8-billion parameter long-context model with a 131,072-token context window. The model achieves 85.37% on HumanEval and 73.84% on MMLU 5-shot, with enhanced tool-calling capabilities reaching 68.27% on BFCL v3. Released under Apache 2.0 license, it supports 12 languages.

Poolside has released Laguna XS.2, a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token, designed for agentic coding. The model features a 131,072-token context window, scores 68.2% on SWE-bench Verified, and is available under Apache 2.0 license with free API access.

Xiaomi released MiMo-V2.5, a 310B total parameter sparse mixture-of-experts model that activates 15B parameters per token. The omnimodal model supports text, image, video, and audio understanding with a 1M token context window and was trained on 48T tokens using FP8 mixed precision.

Xiaomi has released MiMo-V2.5-Pro, an open-source Mixture-of-Experts model with 1.02 trillion total parameters and 42 billion active parameters. The model supports up to 1 million tokens context length and claims 99.6% on GSM8K and 86.2% on MATH benchmarks.

Alibaba's Qwen Team has released Qwen3.6 27B, a 27-billion parameter multimodal language model with a 262,144-token context window. The model accepts text, image, and video inputs and includes a built-in thinking mode for extended reasoning, with pricing at $0.195 per million input tokens and $1.56 per million output tokens.

DeepSeek released a preview of its V4 model on April 24, 2026, claiming the open-source system matches leading closed-source models from Anthropic, Google, and OpenAI. The company emphasized improved coding capabilities and compatibility with domestic Huawei chips, but did not disclose training costs or hardware specifications.

DeepSeek released a preview of its V4 large language model on April 24, 2026, available in 'pro' and 'flash' versions. The Hangzhou-based company claims the open-source model achieves strong performance on agent-based tasks and has been optimized for tools like Anthropic's Claude Code and OpenClaw.

DeepSeek has released DeepSeek-V4-Pro-Base, a 1.6 trillion parameter foundation model now available on Hugging Face. The base model uses BF16 precision for weights and includes support for F8_E4M3, I64, and F32 tensor types.

DeepSeek released two Mixture-of-Experts models: V4-Flash with 284B total parameters (13B activated) and V4-Pro with 1.6T parameters (49B activated). Both models support one million token context windows and use a hybrid attention architecture that requires only 27% of the inference FLOPs compared to DeepSeek-V3.2 at 1M token context.

DeepSeek released two new Mixture-of-Experts models: DeepSeek-V4-Pro with 1.6 trillion parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated), both supporting one million token context length. The models achieve 27% of inference FLOPs and 10% of KV cache compared to DeepSeek-V3.2 at 1M context through a hybrid attention architecture combining Compressed Sparse Attention and Heavily Compressed Attention.

Tencent has released Hy3-preview, a 295-billion-parameter Mixture-of-Experts model with 21 billion active parameters and a 256K context window. The model scores 76.28% on MATH and 34.86% on LiveCodeBench-v6, with particularly strong performance on coding agent tasks.

Open-weight AI models are narrowing the performance gap with closed frontier models in current benchmarks focused on coding and terminal tasks, but industry analysts predict they'll struggle to keep pace as the field shifts toward specialized knowledge work in accounting, law, and healthcare. The gap reduction masks a more complex dynamic where benchmark correlation with real-world performance is weakening.

NVIDIA released GR00T N1.7, a 3-billion parameter open-source Vision-Language-Action model for humanoid robots with commercial licensing. The model was trained on 20,854 hours of human egocentric video data and demonstrates the first documented scaling law for robot dexterity, where increasing human video data from 1,000 to 20,000 hours more than doubles task completion rates.

Tencent has released HY-World 2.0, an open-source multi-modal world model that generates navigable 3D environments from text prompts, single images, multi-view images, or video. The model produces editable 3D assets including meshes and 3D Gaussian Splattings that can be directly imported into game engines like Unity and Unreal Engine.

Baidu has released ERNIE-Image, an 8B parameter text-to-image generation model built on a single-stream Diffusion Transformer architecture. The model is designed for complex instruction following, text rendering, and structured image generation, and can run on consumer GPUs with 24GB VRAM.

The UK's AI Safety Institute published an independent evaluation confirming Anthropic's Claude Mythos is highly effective at finding security vulnerabilities. The evaluation revealed a linear relationship: more tokens spent equals more exploits discovered, transforming security into an economic arms race.

MiniMax has released MiniMax-M2.7, a 229-billion parameter model that uniquely participates in its own evolution during development. The model achieves 66.6% medal rate on MLE Bench Lite and 56.22% on SWE-Pro benchmarks, with native support for multi-agent collaboration and complex tool orchestration.

Google released Gemma 4, an open-source multimodal model that runs entirely on smartphones without sending data to the cloud. The E2B and E4B variants require just 6GB and 8GB of RAM respectively and can autonomously use tools like Wikipedia, maps, and QR code generators through built-in agent skills. The model is available free via the Google AI Edge Gallery app for Android and iOS.

Liquid AI has released LFM2.5-VL-450M, a refreshed 450M-parameter vision-language model built on an updated LFM2.5-350M backbone. The model features a 32,768-token context window, supports 9 languages, handles native 512×512 pixel images, and adds bounding box prediction and function calling capabilities. Performance improvements span both vision and language benchmarks compared to its predecessor.

Tencent has released HY-Embodied-0.5, a family of foundation models designed specifically for embodied AI and robotic control. The suite includes a 2B-parameter MoT (Mixture-of-Transformers) variant with only 2.2B activated parameters during inference, and a 32B model that claims frontier-level performance comparable to Gemini 3.0 Pro, trained on over 200 billion tokens of embodied-specific data.

Meta launched Muse Spark on April 8, 2026, a natively multimodal reasoning model with tool-use and visual chain-of-thought capabilities. Unlike Llama, it is entirely proprietary with no open weights. The model scores 52 on AI Index v4.0 and excels on health benchmarks but represents Meta's departure from its open-source identity.

Google DeepMind has released Gemma 4 26B A4B, a free instruction-tuned Mixture-of-Experts model with 262,144 token context window and multimodal capabilities including text, images, and video input. Despite 25.2B total parameters, only 3.8B activate per token, delivering performance comparable to larger 31B models at reduced compute cost.

Google DeepMind has released Gemma 4 31B Instruct, a free 30.7-billion parameter model with a 256K token context window, multimodal text and image input capabilities, and native function calling. The model supports configurable reasoning mode and 140+ languages, with strong performance on coding and document understanding tasks under Apache 2.0 license.

Microsoft's Bing team has open-sourced Harrier, a 27-billion-parameter embedding model that supports over 100 languages and features a 131,072-token context window. The model ranks first on the MTEB v2 multilingual benchmark, outperforming proprietary offerings from OpenAI and Amazon, and is available on Hugging Face under the MIT license.

Tencent's Hunyuan team released OmniWeaving on April 3, 2026, an open-source video generation model designed to compete with proprietary systems like Seedance-2.0. The model combines multimodal composition, reasoning-informed capabilities, and supports eight video generation tasks including text-to-video, image-to-video, video editing, and compositional generation.

NVIDIA has released a quantized version of Google DeepMind's Gemma 4 31B IT model, compressed to NVFP4 format for efficient inference on consumer GPUs. The 30.7B-parameter multimodal model supports 256K token context windows, handles text and image inputs with video frame processing, and maintains near-baseline performance across reasoning and coding benchmarks.

Google DeepMind released Gemma 4, a multimodal model family supporting text, images, video, and audio with context windows up to 256K tokens. The release includes four sizes (E2B, E4B, 26B A4B, and 31B) designed for deployment from mobile devices to servers. The 31B dense model achieves 85.2% on MMLU Pro and 89.2% on AIME 2026.

Google DeepMind has released Gemma 4, a family of open-weights multimodal models ranging from 2.3B to 31B parameters with support for text, images, video, and audio. The models feature context windows up to 256K tokens, built-in reasoning modes, and native function calling for agentic workflows.

Google DeepMind released the Gemma 4 family of open-source models with multimodal capabilities (text, image, audio, video) and context windows up to 256K tokens. Four distinct model sizes—E2B (2.3B effective parameters), E4B (4.5B effective), 26B A4B (3.8B active), and 31B—are available under the Apache 2.0 license, with instruction-tuned and pre-trained variants.

Google has released Gemma 4, a family of four open models ranging from 2B to 31B parameters, now available under the Apache 2.0 license for the first time. The 31B dense model ranks 3rd on the Arena AI Text Leaderboard, while the 26B mixture-of-experts variant ranks 6th, both outperforming significantly larger competitors. All models support multimodal inputs and are available on Hugging Face, Kaggle, and Ollama.

Google DeepMind has released Gemma 4 31B, a 30.7-billion-parameter multimodal model supporting text and image input. The model features a 262,144-token context window, configurable thinking/reasoning mode, native function calling, and multilingual support across 140+ languages under Apache 2.0 license.

Arcee AI has released Trinity Large Thinking, an open-source reasoning model featuring a 262,144 token context window. The model is priced at $0.25 per million input tokens and $0.90 per million output tokens, with free access available through OpenRouter for the first five days.

Microsoft released the Harrier-OSS embedding model family, comprising three variants with 270M, 600M, and 27B parameters. The largest model achieves 74.3 on the Multilingual MTEB v2 benchmark. All models support 32,768 max tokens and multilingual inputs across 40+ languages.

Microsoft has released Harrier-OSS-v1, a family of multilingual text embedding models trained with contrastive learning and knowledge distillation. The 0.6B parameter variant achieves a 69.0 score on the Multilingual MTEB v2 benchmark with support for 32,768 token context windows and 45+ languages.

Cohere has released Transcribe, a 2 billion parameter open-source automatic speech recognition model that the company claims tops the Hugging Face Open ASR Leaderboard with a 5.42% word error rate. The model supports 14 languages and is available under Apache 2.0 license, outperforming OpenAI's Whisper Large v3 and competing models on both accuracy and throughput metrics.

Chroma has released Context-1, a 20B parameter Mixture of Experts model trained specifically for retrieval tasks that require multi-hop reasoning. The model decomposes complex queries into subqueries, performs parallel tool calls, and actively prunes its own context mid-search—achieving comparable performance to frontier models at a fraction of the cost and up to 10x faster inference speed.

Mistral AI released Voxtral TTS, an open-source text-to-speech model designed for enterprise voice agents and edge devices. The model supports nine languages, adapts custom voices from samples under five seconds, and achieves 90ms time-to-first-audio latency with a 6x real-time factor.

The Allen Institute for AI has released MolmoWeb, a fully open web agent that operates websites using only screenshots without access to source code. The 8B-parameter model achieves 78.2% success on WebVoyager—nearly matching OpenAI's o3 at 79.3%—while being trained on one of the largest public web task datasets ever released.

OpenAI is releasing a set of open-source prompts developers can use to make their applications safer for teens. The policies, designed to work with OpenAI's gpt-oss-safeguard model, address graphic violence, sexual content, harmful body ideals, dangerous activities, and age-restricted goods.

Stability AI has open-sourced Stable Audio Open Small in partnership with Arm, a smaller and faster variant of its text-to-audio model designed for on-device deployment. The model maintains output quality and prompt adherence while reducing computational requirements for real-world edge deployment on devices powered by Arm's technology, which runs on 99% of smartphones globally.

Rakuten Group has released RakutenAI-3.0, a 671 billion parameter mixture-of-experts (MoE) model designed specifically for Japanese language tasks. The model activates 37 billion parameters per token and supports a 128K context window. It is available under the Apache License 2.0 on Hugging Face.

Nvidia has released Nemotron 3 Super, a 120-billion parameter hybrid Mamba-Transformer Mixture-of-Experts model that activates only 12 billion parameters during inference. The open-weight model features a 1-million token context window, multi-token prediction capabilities, and pricing at $0.10 per million input tokens and $0.50 per million output tokens.

NVIDIA released Nemotron 3 Content Safety 4B, an open-source multimodal safety model designed to moderate content across text, images, and multiple languages. Built on Gemma-3 4B-IT with a 128K context window, the model achieved 84% average accuracy on multimodal safety benchmarks and supports over 140 languages through culturally-aware training data.

Cursor acknowledged that its newly launched Composer 2 coding model was built on Moonshot AI's open-source Kimi 2.5 model, not developed independently. The admission came after user scrutiny revealed code references to Kimi. Cursor claims only 25% of compute came from the base model, with 75% from its own training and reinforcement learning.

AI2 has released MolmoSpaces and MolmoBot, robotics models trained exclusively in simulation that transfer directly to real robots without manual real-world data collection or fine-tuning. The approach eliminates months of teleoperated demonstrations typically required for simulation-trained robots. Both systems are open-source.

Hume AI has open-sourced TADA, a speech generation model that maps exactly one audio signal to each text token, achieving 5x faster processing than comparable systems. The model produced zero transcription hallucinations across 1,000+ test samples and runs on smartphones, available in 1B and 3B parameter versions under MIT license.