Explore Text Models

Nvidia's Nemotron 3 Ultra 550B A55B model from the Nemotron 3 family. It uses a hybrid Mamba-Transformer MoE architecture. Provider-specific context limits vary, with the longest current route supporting up to 1M context.

Nvidia's Nemotron 3 Ultra 550B A55B model from the Nemotron 3 family. It uses a hybrid Mamba-Transformer MoE architecture. Provider-specific context limits vary, with the longest current route supporting up to 1M context. Thinking enabled.

Qwen3.6 27B is a dense language model from Alibaba's Qwen team with text and image input, configurable thinking/reasoning behavior, and a native 262K context window. Running inside a TEE (Trusted Execution Environment), with provider attestation support.

Nex AGI's open-source agentic reasoning model, post-trained on Qwen3.5-397B-A17B. It is built for agentic coding, software engineering, deep research, tool use, and long-horizon tasks with a 256K context window.

MiMo V2.5 with Xiaomi thinking enabled. It supports deep reasoning, tool calling, structured outputs, and web search with up to 1M context.

MiMo V2.5 Pro with Xiaomi thinking enabled for coding, long-context reasoning, and agentic orchestration.

Mistral Code Agent Latest is Mistral's direct API alias for devstral-2512, an agentic coding model built for autonomous software engineering, tool use, and long-running code tasks.

Mistral Code Latest is Mistral's direct API alias for codestral-2508, a low-latency coding model for code generation, completion, fill-in-the-middle workflows, function calling, and structured output.

MiniMax M3 is the non-thinking route for MiniMax's open-weights frontier model, built for coding, agent workflows, tool use, and multimodal understanding from step zero. It keeps native thinking disabled for faster direct answers. MiniMax reports 59.0% on SWE-Bench Pro and 66.0% on Terminal Bench 2.1, with Sparse Attention designed to scale context to 1M. It starts with a 512K context cap on NanoGPT for now.

MiniMax M3 Thinking is the adaptive-thinking version of MiniMax's open-weights frontier model for coding, agent workflows, tool use, long-context tasks, and native multimodal understanding. MiniMax reports 59.0% on SWE-Bench Pro and 66.0% on Terminal Bench 2.1, with Sparse Attention designed to scale context to 1M. It starts with a 512K context cap on NanoGPT for now.

Qwen3.7 Plus is Alibaba's cost-effective Qwen 3.7 multimodal agent model for coding, tool use, productivity workflows, visual understanding, screen reading, and GUI interaction.

Qwen3.7 Plus with thinking mode enabled for deeper multimodal reasoning, coding, tool use, screen reading, and productivity workflows.

Step 3.7 Flash Thinking is StepFun's high-efficiency multimodal MoE model with visible reasoning enabled for deeper agentic coding, long-context reasoning, tool use, and native image/video understanding. ⚠️ Note: This model routes through StepFun, so privacy and logging guarantees may be limited.

M-Prometheus 14B is an open multilingual LLM judge from Unbabel, fine-tuned from Qwen2.5 14B to evaluate model outputs with direct assessment, pairwise comparison, and long-form feedback.

Anthropic Claude Opus 4.8 with text, image, and file inputs and a 1M-token context window.

Anthropic Claude Opus 4.8 with thinking enabled.

Qwen3.5 122B A10B is a large MoE model with 122B total parameters, 10B active parameters per token, strong coding and tool-use behavior, and a 262K token context window. Running inside a TEE (Trusted Execution Environment), with provider attestation support.

Gemma 4 31B Instruct is a 30.7B dense model with a long context window, multilingual performance, function calling, and configurable reasoning. Running inside a TEE (Trusted Execution Environment), with provider attestation support.

Qwen3.6 35B A3B Uncensored Aggressive is a native vision-language MoE model tuned to reduce refusal behavior while preserving Qwen3.6 coding, math, and multimodal capabilities. Running inside a TEE (Trusted Execution Environment), with provider attestation support.

Gemma 4 26B A4B Uncensored Heretic is a multimodal MoE model tuned to reduce refusal behavior while preserving coding, reasoning, and function-calling strengths. Running inside a TEE (Trusted Execution Environment), with provider attestation support.

Cohere's open-weight Command A+ model for complex agentic, multimodal, multilingual, and reasoning tasks. It is a 218B total parameter sparse MoE model with 25B active parameters, supports text and image inputs, structured outputs, and 48 languages.

Qwen3.7 Max is Alibaba's latest flagship Qwen model for agentic coding, office automation, and long-running tool workflows in non-thinking mode.

Qwen3.7 Max with thinking mode enabled for deeper reasoning, coding, and long-running tool workflows.

Grok Build 0.1 is xAI's fast coding model trained specifically for agentic software engineering workflows. It supports text and image inputs with text output, and is optimized for interactive coding agents, tool use, and multi-step development tasks. Currently in early access. Content policy rejections can still be charged: xAI may pass through a $0.05 moderation-failure fee, or a $0.055 usage-guidelines violation fee, depending on which rejection upstream returns.

Google's speed-focused Gemini Flash model for frontier multimodal intelligence across text, images, audio, video, PDFs, and code. Built for agentic coding, reliable tool use, structured outputs, and long-context workflows.

Gemini 3.5 Flash with higher thinking enabled for harder reasoning, agentic coding, tool use, multimodal analysis, and long-context workflows.

Perceptron Mk1 is Perceptron's vision-language model for image and video understanding, OCR, document parsing, object detection, counting, spatial localization, and embodied visual reasoning.

Sarvam 30B is a 30B parameter chat completion model from Sarvam AI with multilingual support, streaming, tool calling, reasoning controls, and a 64k context window.

Sarvam 105B is a 105B parameter chat completion model from Sarvam AI with multilingual support, streaming, tool calling, reasoning controls, and a 128k context window.

MiroMind's flagship 235B deep research agent for multi-step research with built-in reasoning and web/tool execution.