Google's Gemma 4 31B instruction-tuned model on Tinfoil's TEE-backed inference path with provider attestation support. This route keeps reasoning disabled for faster direct answers while preserving structured output support.

Google's Gemma 4 26B A4B instruction-tuned model with OpenRouter reasoning enabled, exposing structured reasoning for more deliberate coding, multimodal analysis, and long-context problem solving.

Thinking-enabled GLM 5V Turbo for image, video, and text inputs. Uses the same multimodal foundation model, but keeps OpenRouter reasoning enabled for more deliberate vision-grounded analysis, planning, and tool use. Not included in the subscription.

Qwen 3.6 Plus is a commercial native multimodal agent model with a major step up over the 3.5 series in agentic coding, front-end programming, OCR, object localization, and general vision-language performance. Supports text, image, and video input with a 1M context window.

Google's Gemma 4 26B A4B instruction-tuned model built for scalable reasoning, coding, long-context, and multimodal workflows. This route keeps OpenRouter reasoning disabled for faster direct answers while preserving multimodal and structured output support.

Google's Gemma 4 31B instruction-tuned model for heavier reasoning, coding, agentic workflows, and long-context multimodal understanding. This route keeps tokenizer thinking disabled for faster direct answers.

Google's Gemma 4 31B instruction-tuned model with NVIDIA tokenizer thinking explicitly enabled, exposing reasoning traces for complex multimodal and coding workflows.

Z.ai's native multimodal agent model for vision-based coding and agent workflows. This is the standard non-thinking variant for image, video, and text inputs, tuned for perceive-plan-execute loops, complex coding, and tool-driven task execution. Not included in the subscription.

Open source Arcee reasoning model with a 262K context window, 80K max output, and native reasoning and tool support for agentic workloads.

xAI's Grok 4.20 flagship release with tool calling, multimodal input support, and a 2M-token context window.

Grok 4.20 Multi-Agent is tuned for collaborative agentic workflows while keeping the same 2M-token context window and multimodal support.

Qwen3.5 Omni Flash is Qwen's fast multimodal model. We verified live support for text prompts, images, audio files, and direct video URLs on Alibaba's chat-completions-compatible API. Alibaba describes Flash as a fully evolved version of Qwen3 Omni with audio input support across 60+ languages.

Qwen3.5 Omni Plus is Qwen's stronger general multimodal model. We verified live support for text prompts, images, audio files, and direct video URLs on Alibaba's chat-completions-compatible API. Alibaba describes Plus as a comprehensive evolution of Qwen3 Omni with support for over 10 hours of audio input.

Compatibility alias that routes to the newest version of Claude Sonnet. Currently routes to Claude Sonnet 4.6 Thinking.

Compatibility alias that routes to the newest version of GPT. Currently routes to GPT 5.4.

Compatibility alias that routes to the newest version of Gemini Flash. Currently routes to Gemini 3 Flash Preview Thinking.

Compatibility alias that routes to the newest version of Gemini Pro. Currently routes to Gemini 3.1 Pro Preview High.

Compatibility alias that routes to the newest version of Gemini Flash Lite. Currently routes to Gemini 3.1 Flash Lite Preview.

Compatibility alias that routes to the newest version of Claude Haiku. Currently routes to Claude Haiku 4.5 Thinking.

Compatibility alias that routes to the newest version of Claude Opus. Currently routes to Claude Opus 4.6 Thinking.

Latest high-performance coding model in Kwaipilot's KAT Coder series, built for complex software engineering, SaaS integration, and large-scale production workflows.

GLM-5.1 appears to be an improvement over GLM 5. Not much else is publicly clear yet. It is available as a pay-as-you-go model.

GLM-5.1 with thinking enabled. It appears to be an improvement over GLM 5, but public details are still limited. It is available as a pay-as-you-go model.

Creative Qwen3.5 27B roleplay, story generation, and conversational finetune built on ArliAI's derestricted base with reasoning and vision support.

TheDrummer's Skyfall 31B v4.2 is a roleplay and storytelling model with a 131k context window.

Aion-2.5 is a GLM-5 variant tuned for immersive roleplay and storytelling with stronger tension, conflict, and darker thematic nuance.

MiMo V2 Omni is Xiaomi's frontier omni model for multimodal agent workflows. It natively handles image, video, audio, and text in one shared architecture, with strong grounding, planning, and tool-use behavior.

MiMo V2 Pro is Xiaomi's flagship foundation model for agent systems, built for coding, tool use, orchestration, and long-horizon workflows. It uses a trillion-parameter-class architecture with 42B active parameters and supports up to 1M context.

MiniMax M2.7 is the first model deeply involved in iterating on its own training. It excels in real-world software engineering (SWE-Pro 56.22%), end-to-end project delivery (VIBE-Pro 55.6%), and complex office workflows with strong tool-use compliance and agentic capabilities.

MiniMax M2.7 Turbo is the highspeed and higher priced route for M2.7.