Gemini 2.5 Flash Image Preview is a state of the art image generation model with contextual understanding. It is capable of image generation, edits, and multi-turn conversations.

Grok Code Fast 1 is a speedy and economical reasoning model that excels at agentic coding. With reasoning traces visible in the response, developers can steer Grok Code for high-quality work flows.

DeepSeek-V3.1 is a 671B-parameter hybrid reasoning model (37B active), supporting both "thinking" and "non-thinking" modes via prompt templates. It extends DeepSeek-V3 with two-phase long-context training (up to 128K tokens) and uses FP8 microscaling for efficient inference. The model excels in tool use, code generation, and reasoning, with performance comparable to DeepSeek-R1 but with faster responses. It supports structured tool calling, code agents, and search agents, making it ideal for research and agentic workflows. Successor to DeepSeek V3-0324, it delivers strong performance across diverse tasks.

GPT-5 Chat is tailored for advanced, natural, and context-aware conversations in enterprise environments. It leverages the latest advancements in OpenAI’s conversational AI, supporting multimodal and dynamic dialogue with enhanced context retention and user intent understanding.

The smallest and fastest member of the GPT-5 family, optimized for developer tools, rapid user interactions, and ultra-low latency environments. While it offers limited reasoning depth compared to larger models, GPT-5-Nano preserves essential instruction-following and safety mechanisms. It is the successor to GPT-4.1-nano and is best suited for real-time, cost-sensitive, or embedded applications.

A compact variant of GPT-5, designed for efficient handling of lighter-weight reasoning and conversational tasks. GPT-5 Mini retains the instruction-following and safety features of its larger counterpart, but with reduced latency and cost. It is the direct successor to OpenAI’s o4-mini model, making it ideal for scalable, cost-sensitive deployments.

OpenAI’s most advanced large language model, engineered for high-stakes applications requiring step-by-step reasoning, precise instruction following, and robust code generation. GPT-5 introduces major improvements in factual accuracy, user intent understanding, and hallucination reduction. It supports advanced prompt routing, user-specified intent (such as "think hard about this"), and is optimized for complex workflows in coding, writing, and health-related domains.

OpenAI’s 21B-parameter open-weight Mixture-of-Experts (MoE) model, released under the Apache 2.0 license. Features 3.6B active parameters per forward pass, optimized for low-latency inference and deployability on consumer or single-GPU hardware. Trained in OpenAI’s Harmony response format, it supports reasoning level configuration, fine-tuning, and agentic capabilities such as function calling and structured outputs.

An open-weight, 117B-parameter Mixture-of-Experts (MoE) language model from OpenAI, designed for high-reasoning, agentic, and general-purpose production use cases. Activates 5.1B parameters per forward pass and is optimized for single H100 GPU deployment with native MXFP4 quantization. Supports configurable reasoning depth, full chain-of-thought access, and native tool use, including function calling, browsing, and structured output generation.

Claude Opus 4.1 is Anthropic’s flagship model, offering improved performance in coding, reasoning, and agentic tasks. It achieves high scores on SWE-bench Verified and excels in multi-file code refactoring, debugging precision, and detail-oriented reasoning. The model supports extended thinking up to 64K tokens and is optimized for research, data analysis, and tool-assisted workflows.

Qwen3-235B-A22B-Thinking-2507 is a high-performance, open-weight Mixture-of-Experts (MoE) language model optimized for complex reasoning tasks. Activates 22B of its 235B parameters per forward pass and natively supports up to 262,144 tokens of context. This "thinking-only" variant enhances structured logical reasoning, mathematics, science, and long-form generation, and is instruction-tuned for step-by-step reasoning, tool use, agentic workflows, and multilingual tasks.

Qwen3-Coder-480B-A35B-Instruct is a Mixture-of-Experts (MoE) code generation model developed by the Qwen team. Optimized for agentic coding tasks such as function calling, tool use, and long-context reasoning over repositories. Features 480 billion total parameters, with 35 billion active per forward pass (8 out of 160 experts), and supports variable pricing based on context length.

Qwen3-235B-A22B-Instruct-2507 is a multilingual, instruction-tuned mixture-of-experts language model based on the Qwen3-235B architecture, with 22B active parameters per forward pass. Optimized for general-purpose text generation, including instruction following, logical reasoning, math, code, and tool usage. Supports a native 262K context length and delivers significant gains in knowledge coverage, long-context reasoning, and coding benchmarks.

Kimi K2 Instruct is a large-scale Mixture-of-Experts (MoE) language model developed by Moonshot AI, featuring 1 trillion total parameters with 32 billion active per forward pass. It is optimized for agentic capabilities, including advanced tool use, reasoning, and code synthesis. Kimi K2 excels across a broad range of benchmarks, particularly in coding (LiveCodeBench, SWE-bench), reasoning (ZebraLogic, GPQA), and tool-use (Tau2, AceBench) tasks. It supports long-context inference up to 128K tokens and is designed with a novel training stack that includes the MuonClip optimizer for stable large-scale MoE training.

Grok 4 is xAI’s latest reasoning model, featuring a 256k context window and support for parallel tool calling, structured outputs, and both image and text inputs. Designed for high-throughput, complex reasoning tasks, with pricing that scales for large token requests.

MiniMax-M1 is a large-scale, open-weight reasoning model with 456B total parameters and 45.9B active per token, leveraging a hybrid Mixture-of-Experts (MoE) architecture and a custom "lightning attention" mechanism. It supports context windows up to 1 million tokens and is optimized for long-context understanding, software engineering, agentic tool use, and mathematical reasoning. The model is trained via a custom reinforcement learning pipeline (CISPO) and demonstrates strong performance on FullStackBench, SWE-bench, MATH, GPQA, and TAU-Bench.

A lightweight, high-speed model from xAI, engineered for logic-based tasks that do not require deep domain knowledge. Grok-3-mini is optimized for rapid response and efficient reasoning, making it ideal for applications where speed and concise logic are prioritized over extensive context or specialized expertise. The model exposes raw thinking traces, providing transparency into its decision-making process and enabling advanced debugging or educational use cases.

Grok 3 is xAI’s flagship model, excelling at enterprise use cases such as data extraction, coding, and text summarization. Possesses deep domain knowledge in finance, healthcare, law, and science, and is optimized for high-accuracy, real-world applications.

Gemini 2.5 Flash is Google’s high-performance workhorse model, specifically designed for advanced reasoning, coding, mathematics, and scientific tasks. Includes built-in "thinking" capabilities and is configurable through a "max tokens for reasoning" parameter for fine-tuned performance.

Gemini 2.5 Pro is Google’s state-of-the-art AI model, designed for advanced reasoning, coding, mathematics, and scientific tasks. Employs “thinking” capabilities for nuanced context handling and achieves top-tier performance on multiple benchmarks, including first-place on the LMArena leaderboard.

Magistral is Mistral's first reasoning model, designed for general-purpose use cases that require extended thought processing and high accuracy. It excels in multi-step challenges such as legal research, financial forecasting, software development, and creative storytelling, where transparency and precision are critical.

DeepSeek-R1-0528 is a lightly upgraded release of DeepSeek R1, utilizing more compute and advanced post-training techniques to push its reasoning and inference capabilities to the level of flagship models like O3 and Gemini 2.5 Pro. Excels in math, programming, and logic leaderboards, with a distilled 8B-parameter variant that rivals much larger models on key benchmarks.

Phi-4 is a 14B-parameter model from Microsoft Research, designed for complex reasoning tasks and efficient operation in low-memory or rapid-response scenarios. Trained on a mix of high-quality synthetic and curated data, it is optimized for English language inputs and demonstrates strong instruction following and safety standards. For more details, see the [Phi-4 Technical Report](https://arxiv.org/pdf/2412.08905).

QwQ-32B is the medium-sized reasoning model in the Qwen series, designed for advanced thinking and reasoning tasks. It achieves competitive performance against state-of-the-art models like DeepSeek-R1 and o1-mini, and is particularly strong on hard problems requiring deep analytical skills.

Claude 3.5 Haiku is engineered for real-time applications, delivering quick response times and enhanced capabilities in speed, coding accuracy, and tool use. It is highly suitable for dynamic environments such as chat interactions, immediate coding suggestions, and customer service bots. The model is currently pointing to Claude 3.5 Haiku (2024-10-22).

Claude 3 Opus is Anthropic's most powerful model for highly complex tasks, offering top-level performance, intelligence, fluency, and understanding. It is optimized for advanced research, coding, and multimodal applications, and is benchmarked as a leader in the Claude 3 family.

Claude 3.5 Haiku is Anthropic’s fastest model, featuring enhancements across coding, tool use, and reasoning. It is optimized for high interactivity and low latency, making it ideal for user-facing chatbots, on-the-fly code completions, data extraction, and real-time content moderation. The model does not support image inputs.

Claude 3.5 Haiku is Anthropic’s fastest model, featuring enhancements across coding, tool use, and reasoning. It is optimized for high interactivity and low latency, making it ideal for user-facing chatbots, on-the-fly code completions, data extraction, and real-time content moderation. The model does not support image inputs.

Claude 3.7 Sonnet is an advanced large language model from Anthropic, featuring improved reasoning, coding, and problem-solving abilities. It introduces a hybrid reasoning approach, allowing users to choose between rapid responses and extended, step-by-step processing for complex tasks. The model excels in agentic workflows, front-end development, and full-stack updates, and offers an extended reasoning mode for enhanced accuracy in math, coding, and instruction-following.

Claude Sonnet 4 is a next-generation model from Anthropic, significantly enhancing coding and reasoning capabilities over its predecessor. It achieves state-of-the-art performance on SWE-bench, balances capability and computational efficiency, and is optimized for both routine and complex software development projects. Key features include improved codebase navigation, reduced error rates, and increased reliability in following intricate instructions.

Claude Opus 4.1 is Anthropic’s flagship model, offering improved performance in coding, reasoning, and agentic tasks. It achieves high scores on SWE-bench Verified and excels in multi-file code refactoring, debugging precision, and detail-oriented reasoning. The model supports extended thinking up to 64K tokens and is optimized for research, data analysis, and tool-assisted workflows.

Mistral Large 2 2411 is an updated release of Mistral Large 2, featuring notable improvements in long context understanding, a new system prompt, and more accurate function calling. It is designed for advanced enterprise and research applications requiring high reliability and performance.

Mistral Large 2 (version mistral-large-2407) is Mistral AI’s flagship model, supporting dozens of languages—including French, German, Spanish, Italian, Portuguese, Arabic, Hindi, Russian, Chinese, Japanese, and Korean—and over 80 coding languages. It features a long context window for precise information recall and is optimized for reasoning, code, JSON, and chat tasks.

Mistral Small 3 is a 24B-parameter language model optimized for low-latency performance across common AI tasks. Released under the Apache 2.0 license, it features both pre-trained and instruction-tuned versions for efficient local deployment. The model achieves 81% accuracy on the MMLU benchmark and performs competitively with larger models, while operating at three times the speed on equivalent hardware.

Mistral-Small-3.2-24B-Instruct-2506 is an updated 24B parameter model from Mistral, optimized for instruction following, repetition reduction, and improved function calling. It supports both image and text inputs, delivers strong performance across coding, STEM, and vision benchmarks, and is designed for efficient, structured output generation.

Mistral Medium 3 is a high-performance, enterprise-grade language model that balances state-of-the-art reasoning and multimodal capabilities with significantly reduced operational cost. It excels in coding, STEM reasoning, and enterprise adaptation, and is optimized for scalable deployments across professional and industrial use cases, including hybrid and on-prem environments.

Pixtral Large is a 124B parameter, open-weight, multimodal model built on top of Mistral Large 2. It is capable of understanding documents, charts, and natural images, and is available under both research and commercial licenses. The model is designed for advanced document and image analysis tasks.

Codestral is Mistral’s cutting-edge language model for coding, specializing in low-latency, high-frequency tasks such as fill-in-the-middle (FIM), code correction, and test generation. It is optimized for developer productivity and supports a wide range of programming languages and code-related tasks.

Mistral Saba is a 24B-parameter language model specifically developed for the Middle East and South Asia. It delivers accurate and contextually relevant responses in multiple Indian-origin languages—including Tamil and Malayalam—alongside Arabic. The model is trained on curated regional datasets and is optimized for multilingual and regional applications.

Command A is an open-weights 111B parameter model from Cohere, featuring a 256k context window and optimized for agentic, multilingual, and coding use cases. It delivers high performance with minimal hardware costs, excelling in business-critical workflows that require advanced reasoning, tool use, and language understanding across multiple languages.

Meta’s Llama 3 8B instruct-tuned model, optimized for high-quality dialogue and demonstrating strong performance in human evaluations. Ideal for efficient conversational AI.

Meta’s Llama 3 70B instruct-tuned model, optimized for high-quality dialogue and demonstrating strong performance in human evaluations. Suitable for advanced conversational AI tasks.

Meta’s Llama 3.1 8B instruct-tuned model, designed for fast and efficient dialogue. It performs strongly in human evaluations and is ideal for applications requiring a balance of speed and quality.

Meta’s Llama 3.1 70B instruct-tuned model, optimized for high-quality dialogue use cases. It demonstrates strong performance in human evaluations and is suitable for a wide range of conversational AI applications.

The highly anticipated 400B class of Llama3 is here, offering a 128k context window and impressive evaluation scores. This 405B instruct-tuned version is optimized for high-quality dialogue and demonstrates strong performance compared to leading closed-source models, including GPT-4o and Claude 3.5 Sonnet.

Llama 3.2 1B is a 1-billion-parameter language model focused on efficient natural language tasks, including summarization, dialogue, and multilingual text analysis. Its small size allows for deployment in low-resource environments while maintaining strong performance across eight core languages.

Llama 3.2 3B is a 3-billion-parameter multilingual model optimized for advanced NLP tasks such as dialogue generation, reasoning, and summarization. It supports eight languages and is trained on 9 trillion tokens, excelling in instruction-following, complex reasoning, and tool use.

Llama 3.2 11B Vision is a multimodal model with 11 billion parameters, designed for tasks combining visual and textual data. It excels at image captioning and visual question answering, bridging the gap between language generation and visual reasoning. Pre-trained on a massive dataset of image-text pairs, it is ideal for content creation, AI-driven customer service, and research.

The Meta Llama 3.3 multilingual large language model (LLM) is a pretrained and instruction-tuned generative model with 70B parameters. Optimized for multilingual dialogue, it outperforms many open-source and closed chat models on industry benchmarks. Supported languages include English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.

Llama 4 Scout 17B Instruct (16E) is a mixture-of-experts (MoE) language model from Meta, activating 17 billion parameters out of a total of 109B. It supports native multimodal input (text and image) and multilingual output (text and code) across 12 supported languages. Designed for assistant-style interaction and visual reasoning, Scout uses 16 experts per forward pass and features a context length of 10 million tokens, with a training corpus of ~40 trillion tokens. Built for high efficiency and local or commercial deployment, it is instruction-tuned for multilingual chat, captioning, and image understanding.

Llama 4 Maverick 17B Instruct (128E) is a high-capacity multimodal language model from Meta, built on a mixture-of-experts (MoE) architecture with 128 experts and 17 billion active parameters per forward pass (400B total). It supports multilingual text and image input, and produces multilingual text and code output across 12 supported languages. Optimized for vision-language tasks, Maverick is instruction-tuned for assistant-like behavior, image reasoning, and general-purpose multimodal interaction. Features early fusion for native multimodality and a 1 million token context window.

Llama Guard 4 is a multimodal content safety classifier derived from Llama 4 Scout, fine-tuned for both prompt and response classification. It supports content moderation for English and multiple languages, including mixed text-and-image prompts. The model is aligned with the MLCommons hazards taxonomy and is integrated into the Llama Moderations API for robust safety classification in text and images.

Qwen2.5 72B is the latest in the Qwen large language model series, offering significant improvements in knowledge, coding, and mathematics. It features specialized expert models, improved instruction following, long-text generation (over 8K tokens), structured data understanding, and robust multilingual support for over 29 languages. The model is optimized for resilience to diverse system prompts and enhanced role-play implementation.

Qwen-Turbo is a 1M context model based on Qwen2.5, designed for fast speed and low cost. It is suitable for simple tasks and applications where efficiency and affordability are prioritized over deep reasoning.

WizardLM-2 8x22B is Microsoft AI's most advanced Wizard model, demonstrating highly competitive performance compared to leading proprietary models and consistently outperforming state-of-the-art open-source models. It is an instruct finetune of Mixtral 8x22B and is optimized for complex reasoning and instruction-following tasks. For more information, see the [official release](https://wizardlm.github.io/WizardLM2/).

Mythomax L2 13B is one of the highest performing and most popular fine-tunes of Llama 2 13B, known for its rich descriptive capabilities and roleplay performance. It is widely used in creative and narrative-driven applications.

Qwen-Max is a large-scale Mixture-of-Experts (MoE) model from Qwen, based on Qwen2.5, and provides the best inference performance among Qwen models, especially for complex multi-step tasks. Pretrained on over 20 trillion tokens and further post-trained with curated Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), it is designed for high-accuracy, high-recall applications. The exact parameter count is undisclosed.

Sonar is Perplexity’s lightweight, affordable, and fast question-answering model, now featuring citations and customizable sources. It is designed for companies seeking to integrate rapid, citation-enabled Q&A features optimized for speed and simplicity.

Sonar Pro is an enterprise-grade API from Perplexity, built for advanced, multi-step queries with added extensibility. It can handle longer and more nuanced searches, follow-up questions, and provides double the number of citations per search compared to the standard Sonar model. The model is optimized for large context windows and comprehensive information retrieval.

Sonar Reasoning is a Perplexity model based on DeepSeek R1, designed for long chain-of-thought reasoning with built-in web search. It is uncensored, hosted in US datacenters, and allows developers to leverage extended reasoning for complex queries, making it suitable for research and knowledge-intensive applications.

Sonar Reasoning Pro is Perplexity’s premier reasoning model, powered by DeepSeek R1 with Chain of Thought (CoT) capabilities. Designed for advanced use cases, it supports in-depth, multi-step queries with a larger context window and can surface more citations per search, enabling more comprehensive and extensible responses. Pricing includes Perplexity search costs for integrated web research.

Sonar Deep Research is a research-focused model from Perplexity, engineered for multi-step retrieval, synthesis, and reasoning across complex topics. It autonomously searches, reads, and evaluates sources, refining its approach as it gathers information. This enables comprehensive report generation in domains like finance, technology, health, and current events. The model’s pricing is based on prompt tokens, citation tokens, and the number of searches and reasoning tokens used during its extensive research phase.

xAI’s Grok 2 Vision 1212 is a next-generation vision-language model designed for advanced image-based AI applications. It features robust visual comprehension, refined instruction-following, and strong multilingual support. The model excels at object recognition, style analysis, and visual reasoning, empowering developers to build intuitive, visually aware applications. Enhanced steerability and reasoning capabilities make it a solid foundation for next-generation image solutions. For more details, see the official [xAI announcement](https://x.ai/blog/grok-1212).

Gemini Flash 1.5 8B is optimized for speed and efficiency, delivering enhanced performance in small prompt tasks such as chat, transcription, and translation. Focuses on cost-effective solutions while maintaining high-quality results, making it suitable for real-time and large-scale operations.

Google’s latest multimodal model, supporting both image and video (where available) in text or chat prompts. Optimized for a wide range of language tasks, including code generation, text editing, problem solving, recommendations, and AI agent workflows.

Gemini 2.0 Flash Lite is optimized for extremely fast response times and low cost, while maintaining the quality of larger models. Ideal for real-time and large-scale applications.

Gemini Flash 2.0 offers significantly faster time to first token (TTFT) compared to previous versions, while maintaining quality on par with larger models. Introduces enhancements in multimodal understanding, coding, complex instruction following, and function calling for robust agentic experiences.

DeepSeek V3 is a 685B-parameter, mixture-of-experts model and the latest iteration of the flagship chat model family from the DeepSeek team. Succeeds the previous DeepSeek V3 model and demonstrates strong performance across a variety of tasks.

DeepSeek Prover V2 is a 671B parameter model, speculated to be geared towards logic and mathematics. Likely an upgrade from DeepSeek-Prover-V1.5, but released without an official announcement or detailed documentation.

Qwen3, the latest generation in the Qwen large language model series, features both dense and mixture-of-experts (MoE) architectures to excel in reasoning, multilingual support, and advanced agent tasks. Its unique ability to switch seamlessly between a thinking mode for complex reasoning and a non-thinking mode for efficient dialogue ensures versatile, high-quality performance. The Qwen3-30B-A3B variant includes 30.5 billion parameters (3.3 billion activated), 48 layers, 128 experts (8 activated per task), and supports up to 131K token contexts with YaRN, setting a new standard among open-source models.

Qwen3-14B is a dense 14.8B parameter causal language model from the Qwen3 series, designed for both complex reasoning and efficient dialogue. Supports seamless switching between a "thinking" mode for tasks like math, programming, and logical inference, and a "non-thinking" mode for general-purpose conversation. Fine-tuned for instruction-following, agent tool use, creative writing, and multilingual tasks across 100+ languages and dialects.

Qwen3-32B is a dense 32.8B parameter causal language model from the Qwen3 series, optimized for both complex reasoning and efficient dialogue. Supports seamless switching between a "thinking" mode for tasks like math, coding, and logical inference, and a "non-thinking" mode for faster, general-purpose conversation. Demonstrates strong performance in instruction-following, agent tool use, creative writing, and multilingual tasks across 100+ languages and dialects.

Qwen3-235B-A22B is a 235B parameter mixture-of-experts (MoE) model developed by Qwen, activating 22B parameters per forward pass. Supports seamless switching between a "thinking" mode for complex reasoning, math, and code tasks, and a "non-thinking" mode for general conversational efficiency. Demonstrates strong reasoning ability, multilingual support (100+ languages and dialects), advanced instruction-following, and agent tool-calling capabilities.

Google’s latest open-source multimodal model, Gemma 3 27B, supports vision-language input and text outputs, handles context windows up to 128k tokens, and understands over 140 languages. Offers improved math, reasoning, and chat capabilities, including structured outputs and function calling.

GPT-4.1, a flagship model for advanced instruction following, software engineering, and long-context reasoning. Supports a 1 million token context window and is tuned for precise code diffs, agent reliability, and high recall in large document contexts.

The April 2023 release of GPT-4 Turbo, supporting vision, JSON mode, and function calling. Trained on data up to April 2023, optimized for advanced multimodal tasks.

Preview release of GPT-4, featuring improved instruction following, JSON mode, reproducible outputs, and parallel function calling. Trained on data up to December 2023. Heavily rate-limited while in preview.

The latest GPT-4 Turbo model with vision capabilities, supporting JSON mode and function calling. Trained on data up to December 2023, it is optimized for high-throughput, multimodal applications.

OpenAI’s most advanced small model, GPT-4o mini, supports both text and image inputs with text outputs. It is highly cost-effective, achieving SOTA intelligence and outperforming larger models on key benchmarks, making it ideal for scalable, interactive applications.

GPT-4o (“o” for “omni”) is OpenAI’s latest multimodal model, supporting both text and image inputs with text outputs. Delivers improved performance in non-English languages and visual understanding, while being faster and more cost-effective than previous models.

The August 2024 version of GPT-4o, offering improved structured output capabilities, including support for JSON schema in responses. Maintains high intelligence and efficiency, with enhanced non-English and visual performance.

The November 2024 release of GPT-4o, featuring enhanced creative writing, more natural and engaging responses, and improved file handling. Maintains the intelligence of GPT-4 Turbo while being twice as fast and 50% more cost-effective, with better support for non-English languages and visual tasks.

Specialized GPT-4o variant trained for web search understanding and execution within chat completions, enabling advanced search query comprehension.

Experimental mini version of OpenAI’s o1 model, optimized for STEM tasks with efficient performance. Not recommended for production use and may be heavily rate-limited.

The latest and strongest model family from OpenAI, o1 is designed to spend more time thinking before responding. Trained with large-scale reinforcement learning for chain-of-thought reasoning, it is optimized for math, science, programming, and other STEM tasks, consistently achieving PhD-level accuracy on industry benchmarks.

A cost-efficient language model from OpenAI, optimized for STEM reasoning tasks, especially in science, mathematics, and coding. Supports the `reasoning_effort` parameter for adjustable thinking time and features significant improvements over its predecessor, with better performance on complex questions and lower latency and cost.

A well-rounded, powerful model from OpenAI, setting new standards in math, science, coding, and visual reasoning. Excels at technical writing and instruction-following, and is designed for multi-step problem solving across text, code, and images. BYOK is required for access.

A compact reasoning model in OpenAI’s o-series, optimized for fast, cost-efficient performance while retaining strong multimodal and agentic capabilities. Supports tool use and demonstrates competitive reasoning and coding performance across benchmarks, outperforming its predecessor o3-mini and approaching o3 in some domains. Well-suited for high-throughput scenarios where latency or cost is critical.

The continually updated version of OpenAI ChatGPT 4o, always pointing to the current GPT-4o model used by ChatGPT. Incorporates additional RLHF and may differ from the API version. Intended for research and evaluation, not recommended for production as it may be redirected or removed in the future.

The fastest and most cost-effective model in the GPT-4.1 series, designed for tasks demanding low latency such as classification and autocompletion. Maintains a 1 million token context window and delivers exceptional performance at a small size, outperforming even some larger models on key benchmarks.

A mid-sized GPT-4.1 model delivering performance competitive with GPT-4o at substantially lower latency and cost. Retains a 1 million token context window and demonstrates strong coding ability and vision understanding, making it suitable for interactive applications with tight performance constraints.

A flagship large language model from OpenAI, optimized for advanced instruction following, real-world software engineering, and long-context reasoning. Supports a 1 million token context window and outperforms GPT-4o and GPT-4.5 in coding (54.6% SWE-bench Verified), instruction compliance (87.4% IFEval), and multimodal understanding. Tuned for precise code diffs, agent reliability, and high recall in large document contexts, making it ideal for agents, IDE tooling, and enterprise knowledge retrieval.

A fine-tuned version of o4-mini, specifically optimized for use in Codex CLI. Recommended for code-related tasks, with improved performance in code generation and completion.