SI Nexus Benchmarks

OpenAI latest compact reasoning model optimized for STEM, math, and coding with customizable compute budget per request.

First-of-its-kind open-weights reasoning model utilizing large-scale reinforcement learning to produce long-chain thought traces.

Full-size flagship reasoning model trained with chain-of-thought reinforcement learning across text and visual inputs.

Google DeepMind experimental reasoning model featuring transparent thoughts, 1M context, and high speed at ultra-low price.

Anthropic flagship model widely considered the leading software engineering model for agentic coding and nuanced prose.

A 671B parameter MoE open model requiring only 37B active parameters per token, delivering frontier quality at minimal operational cost.

Alibaba Cloud flagship model built with massive pre-training scale, delivering top international benchmark standings.

Google next-gen workhorse model engineered for low-latency multimodal streaming, function calling, and fast web search integration.

xAI flagship model trained on real-time internet data stream with frontier multimodal capabilities.

OpenAI omni model accepting any combination of text, audio, and image inputs and generating text, audio, and image outputs.

Meta flagship 405 billion parameter open-weights model designed to rival top closed frontier foundation models.

Meta latest 70B open model providing performance rivaling previous generation top-tier models while maintaining modest hardware requirements.

Anthropic fastest model, matching Claude 3 Opus speed while outperforming it across key programming and instruction benchmarks.

Google DeepMind breakthrough long-context model capable of processing up to 2 hours of audio or 1 hour of video in a single prompt.

Mistral AI flagship 123B model designed specifically for high code comprehension and multilingual accuracy.

Open-weights coding champion designed specifically for software engineering that runs locally on modest hardware.

OpenAI lightweight multimodal model replacing GPT-3.5 Turbo with far higher intelligence at a fraction of the price.

Cohere flagship open-weights enterprise model specialized for RAG workflows with multi-step tool use.