Model intelligence
Model information, optional release dates, specifications, official resources, GitHub projects and Hugging Face repositories—organized by model family and vendor.
Google DeepMind’s Gemini models. Gemini 4 Carbon (November 2026) is the most capable model, for deep debugging, architecture, proofs and security audits; Gemini 4 Argon (September 2026) is Google’s recommended default for everyday coding and knowledge work. Both share a 1M-token window and the same per-token pricing, available through the Gemini API, Google AI Studio and Gemini Enterprise.
Anthropic’s Claude models. The current lineup is Claude Fable 5.2 for the hardest reasoning and agent work, Opus 5.5 as the recommended default for coding and knowledge work, Sonnet 5.5 for speed and intelligence, and Haiku 5.5 for high-volume, latency-sensitive tasks. All take text and image input, return text, and have a 1M-token context window.
TypeSafe AI decision model for typed choices, scores and probabilities. No generated prose; input-token billing applies. Evaluate accuracy on your own workload.
Mistral model family from Mistral AI.
DeepSeek R model family from DeepSeek.
DeepSeek V model family from DeepSeek.
35 results
Google DeepMind’s Gemini models. Gemini 4 Carbon (November 2026) is the most capable model, for deep debugging, architecture, proofs and security audits; Gemini 4 Argon (September 2026) is Google’s recommended default for everyday coding and knowledge work. Both share a 1M-token window and the same per-token pricing, available through the Gemini API, Google AI Studio and Gemini Enterprise.
Anthropic’s Claude models. The current lineup is Claude Fable 5.2 for the hardest reasoning and agent work, Opus 5.5 as the recommended default for coding and knowledge work, Sonnet 5.5 for speed and intelligence, and Haiku 5.5 for high-volume, latency-sensitive tasks. All take text and image input, return text, and have a 1M-token context window.
GPT model family from OpenAI.
Gemma model family from Google DeepMind.
DeepSeek V model family from DeepSeek.
Qwen model family from Alibaba Qwen.
Amazon Nova model family from Amazon AGI.
Aya model family from Cohere.
CogView model family from Z.ai / GLM.
Command model family from Cohere.
Cosmos model family from NVIDIA.
DeepSeek R model family from DeepSeek.
ERNIE model family from Baidu.
GLM model family from Z.ai / GLM.
Grok model family from xAI.
Hunyuan model family from Tencent Hunyuan.
Hunyuan3D model family from Tencent Hunyuan.
HunyuanImage model family from Tencent Hunyuan.
HunyuanVideo model family from Tencent Hunyuan.
Jamba model family from AI21 Labs.
Janus model family from DeepSeek.
TypeSafe AI decision model for typed choices, scores and probabilities. No generated prose; input-token billing applies. Evaluate accuracy on your own workload.
Kimi model family from Moonshot AI / Kimi.
Kimi-VL model family from Moonshot AI / Kimi.
Official and editor-reviewed GitHub repositories.
DeepSeek’s own high-performance inference framework, open-sourced for offline inference of DeepSeek-V4.1-Flash on NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs.
Connect Python services to the TypeSafe API using typed questions and results. The repository provides the client library rather than model weights.
Integrate TypeSafe decisions into JavaScript and TypeScript services. Answer types are inferred from questions; the package supports ESM and CommonJS.
Official agent skills for designing applications around TypeSafe typed decisions. Development guidance complements the runtime SDKs and API.
An open-source AI coding agent that lives in your terminal.
Run agents like Hermes, LangChain Deep Agents, and OpenClaw more securely inside NVIDIA OpenShell with managed inference
Lightweight coding agent that runs in your terminal
TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT LLM also contains components to create Python and C++ runtimes that orchestrate the inference execution in a performant way.