AI models to compare before you build around one app

A DeepReinforce model family for agentic coding, with public Hugging Face checkpoints from 9B to 397B scale, MIT license tags, GGUF and FP8 variants, and a company post describing self-scaffolding reinforcement learning and project-reported coding-agent benchmark results.

A Cohere Labs 30B-total, 3B-active mixture-of-experts coding model for code generation, agentic software engineering, and terminal tasks, with public BF16 and FP8 Hugging Face weights, Apache 2.0 licensing, and OpenCode or Cohere API try paths listed by official materials.

A JetBrains 12B MoE model family for software-engineering AI workflows, with 2.5B active parameters per token, multiple public checkpoints, 131,072-token context notes, tool-use and agent-workflow framing, deployment examples, and a technical report.

A Z.ai flagship text-generation model positioned around 1M-token context, long-horizon coding, agentic engineering, public model weights, API access, and local serving paths.

A Moonshot AI coding-focused agentic model built on Kimi-K2.6, with a 256K context length, image and video input notes, preserve-thinking behavior, multi-step tool-call materials, native INT4 quantization, and deployment paths through vLLM, SGLang, KTransformers, and Moonshot API access.

A multimodal agentic model positioned around long-horizon coding, tool use, autonomous execution, and broader software workflows.

An open-weight multimodal model positioned around agentic coding, tool use, long-context work, and real-world software workflows.

A Qwen language world model for simulating agentic environments, with Apache-2.0 Hugging Face weights, a GitHub repo, AgentWorldBench data, prompts, evaluation/deployment scripts, and domains including MCP, search, terminal, software engineering, Android, web, and OS tasks.

A MiniMax native multimodal model with public ModelScope and Hugging Face pages, 1M context, about 428B total parameters and 23B activated parameters, MiniMax Sparse Attention materials, thinking and non-thinking modes, and deployment paths for Transformers, vLLM, and SGLang.

A DeepSeek model family release positioned around long-context intelligence, reasoning modes, coding benchmarks, and agentic task evaluation.

A Tencent Hy Team MoE model positioned around long-context reasoning, instruction following, coding, and agent task evaluation.

A StepFun multimodal MoE model collection with BF16, FP8, NVFP4, and GGUF variants, 256K context notes, tool-use and agent-workflow framing, and deployment paths across vLLM, SGLang, Transformers, and llama.cpp.

A Cohere Labs Command A+ model variant with W4A4 quantization, positioned around agentic tool use, multimodal inputs, multilingual work, long context, and Cohere-reported lower hardware requirements.

A model designed for coherent multi-turn behavior, clean tool use, constrained instruction following, and efficient serving at scale.

A Google DeepMind Gemma 4 model family collection with public checkpoints including Gemma 4 12B, a dense multimodal model Google describes around local agentic workflows, native audio input, and encoder-free vision/audio handling.

Liquid AI's 230M-parameter LFM2.5 text model for lightweight on-device agentic pipelines, data extraction, and edge inference, with Hugging Face weights, GGUF, ONNX, and MLX variants, 32K context notes, tool-use guidance, and official docs/blog materials.

A hybrid model in the LFM2.5 family built for on-device deployment, with extended pre-training and reinforcement learning.

A Liquid AI hybrid model with 8.3B total parameters, 1.5B active parameters, long-context support, tool-use notes, local formats, and deployment paths for edge and agentic workflows.

An OpenBMB 1B-class language model for local assistants, coding agents, tool-use, reasoning, and long-context workflows, with 131K context, Think / No Think modes, standard LlamaForCausalLM architecture, and deployment paths across Transformers, vLLM, SGLang, Docker, GGUF, Ollama, LM Studio, and MLX.

A small Zyphra mixture-of-experts reasoning model with public weights, 760M active parameters, 8.4B total parameters, deployment notes, and project-reported math and coding evaluations.

An NVIDIA 4B content-safety model for classifying prompts, optional images, and model responses against standard or custom policies, with a Hugging Face model card, launch post, released dataset, Transformers, vLLM, SGLang, and NVIDIA NIM paths to inspect.

An OpenMOSS vision-language family with Base and Instruct releases for image, video, OCR, and document understanding work.

A family of vision-language encoders from Google DeepMind, positioned around image-text pretraining, spatial awareness, and general-purpose multimodal applications.

A newer DeepSeek OCR model release for image/PDF OCR, document-to-Markdown workflows, dynamic resolution, vLLM/Transformers inference, and visual causal flow research.

A multimodal OCR model for complex document understanding, positioned around strong real-world document parsing and efficient deployment.

A Baidu OCR model and code release for one-shot long-horizon document parsing, with public GitHub, Hugging Face, ModelScope, and arXiv materials, Transformers and SGLang examples, and batch image/PDF inference paths.

A ByteDance Research unified multimodal model for image and video understanding, generation, and editing, with model files, demos, inference scripts, Gradio setup, benchmark scripts, and a stated 40GB VRAM inference requirement.

A Google Research time-series forecasting foundation model with a public GitHub repo, TimesFM 2.5 model notes, PyPI package, Hugging Face checkpoints, Apache 2.0 licensing, and related Google BigQuery ML documentation for the supported product path.

An NVIDIA Isaac GR00T N1.7 vision-language-action model family for humanoid and generalist robot skills, with a 3B model, post-trained variants, GitHub code, inference and fine-tuning notes, LeRobot-format workflow support, and official robotics developer materials.

An Ai2 remote-sensing foundation model family for satellite imagery and planetary-scale mapping, with v1.1 Base and BandExtractor models, model weights, training code, a technical report, and Ai2-reported lower compute cost.

A low-bit on-device translation model from AngelSlim, positioned around 33-language offline translation, GGUF access, Android demo use, and 1.25-bit compression.