Open-Source Models for Agents

Chinese-optimized open model family with strong domain performance for multilingual agent deployments.

Cohere's 104B parameter model optimized specifically for RAG and enterprise tool use in agent workflows.

A 671B MoE model that delivers GPT-4 level reasoning at 68x lower cost for budget-conscious agent deployments.

Efficient open model from TII UAE with strong multilingual support across 10B to 180B parameter variants.

Efficient multimodal model from Google in 1B to 27B sizes ideal for edge and on-device agents.

Consumer and IoT-optimized model from Google in 2B to 31B sizes with E2B and E4B variants.

Zhipu AI's 744B MoE model achieving the lowest hallucination rate and 77.8% on SWE-bench.

Zhipu AI’s newest open-source GLM family release optimized for large-context reasoning and agentic tool use.

Shanghai AI Lab model with strong long-context reasoning capabilities for complex agent tasks.

Meta's flagship open model family with Scout and Maverick variants and strong tool use capabilities.

European open model at 123B parameters with strong function calling for agent workflows.

Microsoft's compact 14B parameter model delivering strong reasoning for on-device and edge agent deployments.

MCP-native multilingual model family from Alibaba with thinking mode across 0.6B to 235B sizes.

Agentic-focused model with 1M context window, repo-level coding, and MCP-native tool use.

Code-focused open model trained on 600+ programming languages for specialized coding agents.

A multilingual model from 01.AI with long context and fast inference across 6B to 200B parameter sizes.