Stop Choose The Wrong AI Models

Advanced instruction following, software engineering, and long-context reasoning with 1M token context window.

Multimodal model, supports text and image inputs, twice as fast and 50% more cost-effective than GPT-4 Turbo.

Compact reasoning model optimized for tool use, coding, and multimodal capabilities with competitive benchmark performance.

Mid-sized model with GPT-4o performance at lower cost and latency, 1M token context, strong coding abilities.

Lighter version of GPT-4o with state-of-the-art intelligence in a cost-efficient mini-model.

Sets new standards for math, science, coding, and visual reasoning tasks with 200K token context window.

Cost-efficient language model optimized for STEM reasoning tasks, excelling in science, mathematics, and coding.

Designed to spend more time thinking before responding, trained with chain-of-thought reinforcement learning.

Compact version of o1 with thinking capabilities at a more cost-effective price point.

Fastest and cheapest GPT-4.1 model, ideal for classification and autocompletion tasks with 1M context window.

Fine-tuned version of o4-mini specifically optimized for code generation in Codex CLI environments.

World's best coding model, benchmarked for sustained performance on complex, long-running tasks and agent workflows.

Enhanced Sonnet with improved coding, agent reliability, and codebase navigation capabilities with 72.7% SWE-bench performance.

Advanced model with hybrid reasoning, improved coding, and step-by-step processing for complex tasks.

Better-than-Opus capabilities, faster speeds, at Sonnet prices with improved reasoning and coding.

Enhanced speed with improved coding accuracy and tool use, ideal for real-time applications and chat interactions.

Anthropic's most powerful model for highly complex tasks with top-tier performance, intelligence, and fluency.

Balanced intelligence and speed for enterprise workloads, offering maximum utility at a moderate price point.

Fastest and most compact Claude model for near-instant responsiveness with quick and accurate targeted performance.

Advanced capabilities with 200K token context window, reduced hallucination rates, and tool use capabilities.

DeepSeek: R1

671B (37B active)

Open-source large model, matches OpenAI o1's reasoning, STEM, math and code performance; 671B parameters, 37B active/step.

DeepSeek: V3 (Chat/0324)

685B
(Mixture-of-Experts)

Latest flagship MoE model. Excels at chat, reasoning, math, code, and general dialog. Up to 128K tokens context.

R1 Distill Qwen 14B

14B

Distilled dense model, beats o1-mini in math/coding. AIME 2024 pass@1: 69.7, MATH-500: 93.9.

DeepSeek: Prover V2

671B

Upgraded logic/mathematics MoE model. Focuses on rigorous mathematical and logical reasoning.

DeepSeek: V3 Base

671B (MoE)

Pretrained V3 base (671B), strong performance on language, math, and coding, supports huge input length.

State-of-the-art AI model for advanced reasoning, coding, mathematics and scientific tasks with "thinking" capabilities.

Workhorse model designed for advanced reasoning with faster response times and built-in thinking capabilities.

Latest multimodal model with 2M token context, supporting image and video understanding in text prompts.

Efficient multimodal model for visual understanding, classification, summarization, and content creation.

Optimized for speed and efficiency, excellent for chat, transcription, and translation with reduced latency.

Google's latest open source model with multimodal capabilities, 128k context, and support for over 140 languages.

Mid-sized multimodal model with vision-language input, multilingual support, and structured outputs.

Compact multimodal model with vision capabilities, supporting structured outputs and function calling.

Smallest Gemma 3 model with 32k context, 140+ languages support, and improved reasoning capabilities.

Open model built from the same research and technology used to create Gemini models.

Advanced open-source language model setting new standards for efficiency and performance in its size class.

Faster TTFT compared to previous versions with quality matching larger models at economical prices.

Multimodal, advanced math and code, fast edge/local inference, 128K context.

Mixture-of-Experts 141B model, SOTA cost/efficiency, powerful code/math/reasoning.

Mixture-of-Experts model, excels at "thinking" mode for advanced reasoning, math and code; supports 131K context, multilingual.

Dense 32.8B model, balances reasoning and multi-language chat, 32K native/131K context.

State-of-the-art multimodal vision-language, RLHF-finetuned, excels at visual-text understanding and video analysis.

The latest general-purpose workhorse, improving code, math, instruction following, and structured data capabilities, with 128K long context.

Latest xAI model designed for data extraction, coding, and deep domain knowledge. Excels at math, science, and structured tasks.

Lightweight, “thinks before responding,” shines on math and reasoning-heavy tasks.

Experimental vision-language model for advanced visual comprehension and reasoning tasks.

Multilingual SOTA model for powerful dialogue, code, and multilingual applications.

Multimodal vision-language model, early fusion, up to 1M-token context, supports 12 languages.

Multimodal content safety classification for both prompt and response, supports text/image.

Enhanced for math, science, and code reasoning, with step-by-step answers, ideal for structured logic workflows.

Compact model for advanced reasoning and instruction, supports text and visual input, runs efficiently on edge/mobile.

Post-trained DeepSeek-R1 variant with improved responsiveness on previously blocked topics and enhanced safety.

Advanced reasoning with DeepSeek R1 + chain-of-thought, large context, multi-citation, ideal for in-depth queries.

Extensible API for nuanced searches, double citation capacity, and longer context vs. baseline Sonar.

Autonomous retrieval, synthesis, multi-query research—ideal for finance/tech/health in-depth reports.