TriOngle

Platforms we work with

The models, agent frameworks, managed runtimes and protocols in use today, named in plain text. Our work is vendor-neutral: we assess the system you run, on whichever of these it is built.

Every name on this page is a trademark of its owner and is used only to identify the platform. Listing a platform implies no partnership, endorsement, sponsorship or certification, and we show no third-party logos. Model names and versions were checked against each provider’s own pages on 4 November 2026; they change often, so every report records the exact versions in scope.

LLM families 19

  • OpenAIGPT modelsGPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna.⁠[1]
  • AnthropicClaude modelsClaude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5.⁠[2]
  • GoogleGemini and Gemma modelsGemini 4 Argon and the Gemini 3 and 2.5 lines;⁠[3] Gemma 4 open models.⁠[4]
  • MetaMuse and Llama modelsMuse Wonder, Muse Spark 1.3, Muse Spark Open and Muse Glimmer through the Meta Model API; Llama.⁠[5]
  • Mistral AIMistral modelsMistral Large 4, Mistral Small 4, Ministral 3 and Codestral; Small 4 and Ministral 3 are open-weight under Apache 2.0.⁠[6]
  • xAIGrok modelsGrok 4.7 flagship.⁠[7]
  • DeepSeekDeepSeek modelsDeepSeek V4 Pro and DeepSeek Flash through its API.⁠[8]
  • Alibaba CloudQwen modelsQwen3.8 models and the Qwen3Guard safety models.⁠[9]
  • Moonshot AIKimi modelsKimi K3.1 Pro.⁠[10]
  • Z.aiGLM modelsGLM-5.3 and GLM-5.3-Flash.⁠[11]
  • CohereCommand, Aya, Embed and RerankModels for chat, tool-using agents, retrieval, multilingual work and search.⁠[12]
  • AmazonAmazon Nova modelsNova 2 Lite and Nova 2 Sonic.⁠[13]
  • Microsoft AIMAI modelsMAI-Thinking-1, MAI-Code-1.1-Flash and the MAI voice, transcription and image models.⁠[14]
  • IBMGranite modelsGranite 4.2 and the Granite Guardian guardrail models, under Apache 2.0.⁠[15]
  • NVIDIANemotron modelsNemotron 3 and 3.5 open models, with open weights and training data.⁠[16]
  • AI21Jamba modelsHybrid Transformer-Mamba models, open weights under the Jamba Open Model License.⁠[17]
  • Nous ResearchHermes modelsHermes 4 models.⁠[18]
  • MiniMaxMiniMax M modelsMiniMax M3 and M3.1 Flash.⁠[19]
  • XiaomiMiMo modelsMiMo V2.5 and V2.6 models.⁠[20]

Agent frameworks 15

Latest release as listed on the project’s page on 4 November 2026.

  • OpenAI Agents SDKOpenAI · MIT · 0.24.0, 15 Oct 2026A lightweight toolkit for multi-agent orchestration that works with many model providers.⁠[21]
  • Agent Development Kit (ADK)Google · Apache 2.0 · 2.13.0, 23 Oct 2026A code-first Python framework for building, evaluating and deploying agents.⁠[22]
  • Microsoft Agent FrameworkMicrosoft · MIT · 1.24.0, 30 Oct 2026A framework for building AI agents in Python.⁠[23]
  • LangGraphLangChain · MIT · 1.2.19, 30 Oct 2026A low-level framework for stateful, multi-actor, long-running agents.⁠[24]
  • LangChainLangChain · MIT · 1.4.8, 2 Nov 2026Composable components for building LLM applications.⁠[25]
  • CrewAIMIT · 1.15.35, 3 Nov 2026A framework for orchestrating role-playing, autonomous agents.⁠[26]
  • LlamaIndexMIT · 0.14.25, 21 Sep 2026A framework connecting LLMs and agents to your data.⁠[27]
  • Haystackdeepset · Apache 2.0 · 3.4.1, 21 Oct 2026A framework for customisable, production-ready LLM applications.⁠[28]
  • smolagentsHugging Face · 1.26.0, 29 May 2026A minimal library in which agents write Python code to call tools or other agents.⁠[29]
  • Pydantic AIPydantic · MIT · 2.65.0, 3 Nov 2026A type-safe agent framework.⁠[30]
  • MastraMastra · Apache 2.0, with an enterprise licence for some directoriesA TypeScript framework for AI applications and agents.⁠[31]
  • OpenHandsOpenHands · MITA self-hosted control centre for coding agents, with optional Docker sandboxing.⁠[32]
  • Strands AgentsAWS · Apache 2.0 · 1.63.0, 30 Oct 2026A model-driven approach to building agents in a few lines of code.⁠[33]
  • Hermes AgentNous Research · MITA self-improving agent that creates skills from experience, with command approval and container isolation.⁠[34]
  • OpenRigIndividual maintainer · Apache 2.0An open-source harness for persistent teams of coding agents; its README warns of broad default permissions.⁠[35] See AG3 Coding-agent and CI hardening.

Managed agent runtimes 3

  • Amazon Bedrock AgentCoreAWSManaged agent infrastructure with a Gateway that enforces Policy, written in natural language and converted to Cedar; Policy generally available since 3 March 2026.⁠[36]
  • Gemini Enterprise Agent Platform, Agent RuntimeGoogle CloudAgent Runtime with an Agent Gateway that can apply Model Armor guardrails, and an Agent Identity API in preview.⁠[37]
  • Microsoft Foundry Agent ServiceMicrosoftA managed platform for building, deploying and scaling agents, with a Microsoft Entra identity per agent and virtual-network isolation;⁠[38] Entra Agent ID is generally available.⁠[39]

See AG9 Managed agent runtime review for how we review a managed runtime.

Protocols and open formats 4

  • Model Context Protocol (MCP)Agentic AI Foundation · specification 2026-07-28An open protocol connecting LLM applications to data sources and tools.⁠[40][41]
  • Agent2Agent (A2A)Linux Foundation · Apache 2.0 · v1.0.1, May 2026An open protocol for agents built on different frameworks to communicate and collaborate.⁠[42][43]
  • AGENTS.mdAgentic AI FoundationA simple, open format for guiding coding agents: a README written for agents.⁠[44]
  • Agent SkillsAgentic AI FoundationAn open format for reusable agent expertise.⁠[41]

Not on the list?

If your system runs on a model, framework or runtime that is not listed, ask us. The method does not depend on any one platform.

References

Numbered in the order they are first cited on this page; each entry links back to the places that cite it. Every source was opened on 4 November 2026.

  1. ↑ Models. OpenAI API documentation.
  2. ↑ Models overview. Claude Platform documentation.
  3. ↑ Models. Google AI Studio documentation.
  4. ↑ Gemma. Google DeepMind.
  5. ↑ Products & Solutions for AI Developers. Meta.
  6. ↑ Models Overview. Mistral Docs.
  7. ↑ Grok Models & Pricing. docs.x.ai.
  8. ↑ Your First API Call. DeepSeek API Docs.
  9. ↑ Qwen. Hugging Face.
  10. ↑ Main Concepts. Kimi API Platform documentation.
  11. ↑ Quick Start. Z.AI Open Platform.
  12. ↑ An Overview of Cohere’s Models. Cohere documentation.
  13. ↑ Amazon Nova foundation models. AWS.
  14. ↑ Models. Microsoft AI.
  15. ↑ Granite. IBM.
  16. ↑ Nemotron AI Models. NVIDIA Developer.
  17. ↑ Jamba: Hybrid Transformer-Mamba Models. AI21.
  18. ↑ NousResearch. Hugging Face.
  19. ↑ MiniMax. minimax.io.
  20. ↑ Xiaomi MiMo. Hugging Face.
  21. ↑ openai-agents. PyPI.
  22. ↑ google-adk. PyPI.
  23. ↑ agent-framework. PyPI.
  24. ↑ langgraph. PyPI.
  25. ↑ langchain. PyPI.
  26. ↑ crewai. PyPI.
  27. ↑ llama-index. PyPI.
  28. ↑ haystack-ai. PyPI.
  29. ↑ smolagents. PyPI.
  30. ↑ pydantic-ai. PyPI.
  31. ↑ mastra-ai/mastra. GitHub.
  32. ↑ OpenHands/OpenHands. GitHub.
  33. ↑ strands-agents. PyPI.
  34. ↑ NousResearch/hermes-agent. GitHub.
  35. ↑ mvschwarz/openrig. GitHub. README, read 4 November 2026.
  36. ↑ Policy in Amazon Bedrock AgentCore is now generally available. AWS, 3 March 2026.
  37. ↑ Gemini Enterprise Agent Platform release notes. Google Cloud Documentation.
  38. ↑ Foundry Agent Service overview. Microsoft Learn.
  39. ↑ What’s new in Microsoft Entra Agent ID. Microsoft Learn, updated 1 May 2026.
  40. ↑ Model Context Protocol: Specification. modelcontextprotocol.io. Schema version 2026-07-28.
  41. ↑ a b Agentic AI Foundation. aaif.io.
  42. ↑ a2aproject/A2A. GitHub.
  43. ↑ A2A releases. GitHub. v1.0.1, May 2026.
  44. ↑ AGENTS.md. agents.md.