Reasoning, coding, OCR, long context, and multimodal tasks reward different model families.
Open models by workload and deployment path
Browse models by the work they are likely to power: reasoning, coding, multimodal understanding, OCR, local inference, and self-hosted agent stacks.
Open weights, hosted API access, local runtimes, and self-hosting change cost and data control.
Use your own agent traces, latency budget, and license constraints before trusting public rankings.
Reasoning and coding models
Models to test for planning, code changes, technical review, and tool-heavy agent prompts.
DeepSeek-R1
Open reasoning model family for developers testing long-form reasoning, coding, and local AI workflows.
Open sourceLocal firstSelf-hosted
Qwen3.5
Alibaba's flagship open model with 397B-A17B MoE architecture, 8.6× decoding improvement over Qwen3, multimodal, 256K context.
Open sourceLocal firstSelf-hosted
Phi-4
Microsoft's compact 14B dense reasoning model, MIT-licensed, tops MMLU in its size class with 16K context.
Open sourceLocal firstSelf-hosted
Mistral Large 3
Europe's most powerful open model, 675B MoE (41B active), agentic-tuned, strong multilingual performance.
Open sourceLocal firstSelf-hosted
Llama 4
Meta's flagship open MoE model family with Scout (109B, 10M context) and Maverick (400B, rivaling GPT-5 on coding).
Open sourceLocal firstSelf-hosted
Rapid-MLX
Apple Silicon local AI engine with OpenAI-compatible API, tool calling, prompt cache, and MLX acceleration.
Open sourceLocal first
Gemma 4 12B
Google DeepMind's 12B open multimodal model for local agentic workflows on laptops.
Open sourceLocal firstSelf-hosted
DeepSeek V4
Open DeepSeek V4 model family for million-token context, coding, reasoning, and agent workflows.
Open sourceLocal firstSelf-hostedAPI
GLM-5
Open model line from Z.ai focused on agentic engineering and longer coding workflows.
Open source
GLM-OCR
Open OCR model and pipeline for turning complex document images into usable text.
Open source
Qwen3-VL
Open vision-language model family for images, screens, documents, and multimodal workflows.
Open source
Qwen3.6
Qwen's open model line focused on stronger coding, agentic tasks, and real-world stability.
Open source
Kimi K2.5
Moonshot AI's open-weight multimodal model for agentic and tool-using workflows.
Multimodal, OCR, and document models
Models for image understanding, documents, screenshots, OCR, and visual agent workflows.
Gemma 4
Google DeepMind's open model family for local, multimodal, and agentic AI workflows.
Open sourceLocal firstSelf-hostedAPI
Gemma 4 12B
Google DeepMind's 12B open multimodal model for local agentic workflows on laptops.
Open sourceLocal firstSelf-hosted
GLM-OCR
Open OCR model and pipeline for turning complex document images into usable text.
Open source
Qwen3-VL
Open vision-language model family for images, screens, documents, and multimodal workflows.
Open source
Local and self-hosted models
Open-weight candidates for teams that care about privacy, reproducibility, and deployment control.
DeepSeek-R1
Open reasoning model family for developers testing long-form reasoning, coding, and local AI workflows.
Open sourceLocal firstSelf-hosted
gpt4all
Run large language models locally on consumer hardware with a desktop application and Python library.
Open source
Qwen3.5
Alibaba's flagship open model with 397B-A17B MoE architecture, 8.6× decoding improvement over Qwen3, multimodal, 256K context.
Open sourceLocal firstSelf-hosted
FinGPT
Open-source financial large language model project for finance sentiment, analysis, and domain adaptation.
Open source
Phi-4
Microsoft's compact 14B dense reasoning model, MIT-licensed, tops MMLU in its size class with 16K context.
Open sourceLocal firstSelf-hosted
Mistral Large 3
Europe's most powerful open model, 675B MoE (41B active), agentic-tuned, strong multilingual performance.
Open sourceLocal firstSelf-hosted
Llama 4
Meta's flagship open MoE model family with Scout (109B, 10M context) and Maverick (400B, rivaling GPT-5 on coding).
Open sourceLocal firstSelf-hosted
OLMo 2
Fully open language model family from AI2 for transparent research, training, and evaluation.
Open sourceLocal firstSelf-hosted
LiteRT-LM
Google's open-source inference framework for deploying large language models on edge devices.
Open sourceLocal first
Rapid-MLX
Apple Silicon local AI engine with OpenAI-compatible API, tool calling, prompt cache, and MLX acceleration.
Open sourceLocal first
Gemma 4
Google DeepMind's open model family for local, multimodal, and agentic AI workflows.
Open sourceLocal firstSelf-hostedAPI
Gemma 4 12B
Google DeepMind's 12B open multimodal model for local agentic workflows on laptops.
Open sourceLocal firstSelf-hosted
DeepSeek V4
Open DeepSeek V4 model family for million-token context, coding, reasoning, and agent workflows.
Open sourceLocal firstSelf-hostedAPI
GLM-5
Open model line from Z.ai focused on agentic engineering and longer coding workflows.
Open source
GLM-OCR
Open OCR model and pipeline for turning complex document images into usable text.
Open source
Mistral Small 3.2
Apache-licensed small open model for practical instruction following, local inference, and agent experiments.
Open sourceLocal firstSelf-hostedAPI
Qwen3-VL
Open vision-language model family for images, screens, documents, and multimodal workflows.
Open source
Qwen3.6
Qwen's open model line focused on stronger coding, agentic tasks, and real-world stability.
Open source
Kimi K2.5
Moonshot AI's open-weight multimodal model for agentic and tool-using workflows.
Compact and general-purpose candidates
Useful baselines for cheaper inference, edge experiments, and broad assistant workloads.
Mistral Large 3
Europe's most powerful open model, 675B MoE (41B active), agentic-tuned, strong multilingual performance.
Open sourceLocal firstSelf-hosted
OLMo 2
Fully open language model family from AI2 for transparent research, training, and evaluation.
Open sourceLocal firstSelf-hosted
Gemma 4
Google DeepMind's open model family for local, multimodal, and agentic AI workflows.
Open sourceLocal firstSelf-hostedAPI
Gemma 4 12B
Google DeepMind's 12B open multimodal model for local agentic workflows on laptops.
Open sourceLocal firstSelf-hosted
Mistral Small 3.2
Apache-licensed small open model for practical instruction following, local inference, and agent experiments.
Open sourceLocal firstSelf-hostedAPI