GPT-6 AstraOpenAIBuilt for hard, end-to-end work: reasoning, coding, computer use and research in one task.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.Gemini 3.8 FlashGoogleFast and cheap, built for long software tasks, agents and multi-step reasoning.Supports reasoning, web search; takes text, image, gives text.DeepSeek V4 ProDeepSeekDeepSeek's most capable model: chat, code, hard problems and multi-step work. Deeper reasoning, and it costs more.Supports reasoning, web search; takes text, gives text.Grok 4.5xAIStrong at code, maths and technical problems, and work that runs over many steps.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.Claude Fable 5AnthropicHandles hard problems, long documents, and work that runs over many steps. Claude Fable 5.1 is newer.Supports reasoning, web search; takes text, image, gives text.Claude Opus 5AnthropicStrong at complex code and work that runs over many steps.Supports reasoning, web search; takes text, image, gives text.Gemini 3.6 FlashGoogleFast, capable and cheap across chat, code and multi-step work. Gemini 3.8 Flash is newer.Supports reasoning, web search; takes text, image, gives text.GPT-5.6 SolOpenAIHandles hard problems, long documents, and work that runs over many steps. GPT-6 Astra is newer.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.Gemini 3.5 Flash LiteGoogleBuilt for straightforward conversation at high volume. Lighter and cheaper than Flash.Supports reasoning, web search; takes text, image, gives text.GPT-5.6 TerraOpenAIGood at conversation, writing and code. Cheaper than Sol, but not built to run long multi-step jobs.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.Grok 4.3xAIA fast all-rounder for chat, code and hard problems. Reads longer documents than 4.5, but isn't built for long multi-step jobs.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.GPT-5.6 LunaOpenAIBuilt for conversation. The fastest and cheapest GPT-5.6, for everyday questions.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.Grok 4.20 ReasoningxAIAn earlier Grok that thinks before answering, so it is slower.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.Gemini 3.1 Flash ImageGoogleCreates and edits images from a description. Fast, good quality.Supports web search, image generation; takes text, image, gives text, image.Grok 4.20 Non-ReasoningxAIAn earlier Grok that answers immediately, without thinking first.Supports web search, image generation with a tool; takes text, image, gives text.Gemini 3.1 Flash Lite ImageGoogleThe fastest and cheapest way to create and edit images. Less detail.Supports reasoning, image generation; takes text, image, gives text, image.Claude Sonnet 5AnthropicFast and capable across chat, code and multi-step work. A good default for everyday use.Supports reasoning, web search; takes text, image, gives text.Gemini 3 Pro ImageGoogleThe highest-quality images, and the best at getting text right inside the picture. Slower.Supports web search, image generation; takes text, image, gives text, image.GPT Image 2OpenAICreates and edits images from a description.Supports image generation; takes text, image, gives image.Qwen 3.5 0.8BAlibabaLocal · 1.59 GBLightest Qwen 3.5 (~1.6 GB VRAM). Low-resource.Supports reasoning; takes text, gives text.Qwen 3.5 2BAlibabaLocal · 2.19 GBNeeds ~2.2 GB VRAM. Not WebLLM low-resource.Supports reasoning; takes text, gives text.Qwen 3.5 4BAlibabaLocal · 3.78 GBNeeds ~3.8 GB VRAM. Not WebLLM low-resource.Supports reasoning; takes text, gives text.Qwen 3.5 9BAlibabaLocal · 6.28 GBLargest local pick (~6.3 GB VRAM). Stronger, not low-resource.Supports reasoning; takes text, gives text.Llama 3.2 1BMetaLocal · 0.86 GBSmallest local pick (~0.9 GB VRAM). Low-resource. English drafts, not heavy work.Takes text, gives text.Llama 3.2 3BMetaLocal · 2.21 GBStronger than the 1B (~2.2 GB VRAM). Still WebLLM low-resource.Takes text, gives text.Qwen 2.5 1.5BAlibabaLocal · 1.59 GBBetter than Llama 1B on non-English (~1.6 GB VRAM). Low-resource.Takes text, gives text.Llama 3.1 8BMetaLocal · 4.88 GBLargest Llama pick (~4.9 GB VRAM). Not WebLLM low-resource.Takes text, gives text.DeepSeek V4.1 FlashDeepSeekFast and cheap for conversation and multi-step work, and it reads images.Supports reasoning; takes text, image, gives text.GPT Image 2.5 SunburstOpenAICreates and edits images, for work where precise edits matter most.Supports image generation; takes text, image, gives image.GPT Image 2.5 FlareOpenAICreates and edits images quickly, for everyday use.Supports image generation; takes text, image, gives image.Claude Fable 5.1AnthropicFor hard problems, long documents, and work that runs over many steps.Supports reasoning, web search; takes text, image, gives text.Gemini 3.7 FlashGoogleFast and cheap for agents, code and video, over long documents.Supports reasoning, web search; takes text, image, gives text.Grok 4.6xAIBuilt for code, long-running agents and visual work, over long documents.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.Claude Opus 4.8AnthropicStrong at code and detailed technical work, over very long documents.Supports reasoning, web search; takes text, image, gives text.Gemini 3.5 FlashGoogleFast and cheap across chat, code and multi-step work.Supports reasoning, web search; takes text, image, gives text.Gemini 3.1 Flash LiteGoogleAn earlier lightweight Gemini for straightforward conversation.Supports reasoning, web search; takes text, image, gives text.GPT-5.5OpenAIA capable all-rounder: conversation, code, hard problems and multi-step work.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.GPT-5.5 ProOpenAIThe newest of the Pro models. Thinks for longer before answering, over very long documents; built for hard problems, not chat or code.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.Claude Opus 4.7AnthropicCareful, thorough reasoning over very long documents.Supports reasoning, web search; takes text, image, gives text.Gemma 4 31BGoogleDense Gemma 4. Stronger than 26B A4B on coding and reasoning. Understands images.Supports reasoning, web search; takes text, image, gives text.Gemma 4 26B A4BGoogleGemma 4 mixture-of-experts. Faster than 31B (3.8B active). Understands images.Supports reasoning, web search; takes text, image, gives text.GPT-5.4 MiniOpenAIA small, fast, cheaper all-rounder — chat, code and multi-step work.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.GPT-5.4 NanoOpenAIThe cheapest GPT-5.4. Built for straightforward conversation, nothing heavier.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.GPT-5.4OpenAIA cheaper all-rounder for conversation, code and multi-step work. Reads very long documents.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.GPT-5.4 ProOpenAIThinks for longer before answering, over very long documents. Built for hard problems, not chat or code.Supports reasoning, web search, image generation with a tool; takes text, image, gives text.GPT-5.3 CodexOpenAIBuilt for programming: writing, reviewing and fixing code, and working through long tasks on its own. Not meant for general chat.Supports reasoning, web search; takes text, image, gives text.Claude Sonnet 4.6AnthropicBalanced speed and quality for writing and code, over very long documents.Supports reasoning, web search; takes text, image, gives text.Claude Opus 4.6AnthropicHandles chat, code and hard problems over very long documents.Supports reasoning, web search; takes text, image, gives text.GPT Image 1.5OpenAIAn earlier image model. Creates and edits images from a description.Supports image generation; takes text, image, gives text, image.GPT-5.2OpenAIThe same ground as GPT-5.1, with finer control over how long it thinks.Supports reasoning, web search; takes text, image, gives text.GPT-5.2 ProOpenAIThinks for longer before answering. Built for hard problems, not chat or code.Supports reasoning, web search; takes text, image, gives text.Claude Opus 4.5AnthropicAn earlier Opus for chat, code and hard problems. Reads a fifth of what 4.6 and later hold.Supports reasoning, web search; takes text, image, gives text.GPT-5.1OpenAIA general-purpose GPT for conversation, code and multi-step work.Supports reasoning, web search; takes text, image, gives text.Claude Haiku 4.5AnthropicThe fastest and cheapest Claude. Built for short questions and quick replies.Supports reasoning, web search; takes text, image, gives text.GPT Image 1 MiniOpenAIA cheaper, lower-detail version of GPT Image 1.Supports image generation; takes text, image, gives image.GPT-5 ProOpenAIThinks hard on every question, with no dial to turn it down. Built for hard problems, not chat or code, and can write very long answers.Supports reasoning; takes text, image, gives text.Gemini 2.5 Flash ImageGoogleAn earlier image model. Creates and edits images, over a much shorter context than newer ones.Supports image generation; takes text, image, gives text, image.Claude Sonnet 4.5AnthropicAn earlier Sonnet for writing and code. Reads a fifth of what 4.6 and later hold.Supports reasoning, web search; takes text, image, gives text.GPT-5 MiniOpenAIA smaller GPT-5 for conversation and code.Supports reasoning, web search; takes text, image, gives text.GPT-5 NanoOpenAIThe smallest GPT-5. Built for straightforward conversation.Supports reasoning, web search; takes text, image, gives text.GPT-5OpenAIThe first GPT-5: conversation, code, hard problems and multi-step work.Supports reasoning, web search; takes text, image, gives text.GPT Image 1OpenAIAn earlier image model. Creates images from a description, and edits images you upload.Supports image generation; takes text, image, gives image.o3OpenAIA reasoning model that works through problems step by step. Built for hard problems, not chat or code.Supports reasoning, web search; takes text, image, gives text.o4 MiniOpenAIA small, fast reasoning model that works through problems step by step.Supports reasoning, web search; takes text, image, gives text.GPT-4.1OpenAIAnswers straight away without thinking first. Good for conversation and code when speed matters more than depth.Supports web search; takes text, image, gives text.GPT-4.1 MiniOpenAIA smaller, faster GPT-4.1 for straightforward conversation.Supports web search; takes text, image, gives text.GPT-4.1 NanoOpenAIThe smallest GPT-4.1. Built for simple, high-volume conversation.Takes text, image, gives text.o3 MiniOpenAIA smaller, cheaper step-by-step reasoning model. Text only — it can't read images.Supports reasoning; takes text, gives text.o1OpenAIAn early reasoning model: it works through problems step by step. Built for hard problems, not chat or code.Supports reasoning; takes text, image, gives text.GPT-4o MiniOpenAIA small, cheap conversation model for short, narrow tasks.Supports web search; takes text, image, gives text.GPT-4oOpenAIA conversation model that answers straight away. Handles far shorter documents than newer GPTs.Supports web search; takes text, image, gives text.GPT-3.5 TurboOpenAIThe oldest GPT here. Short conversations only — it holds a fraction of what any newer model does.Takes text, gives text.
Takes gives TextImagesCan doReasoningWeb searchMakes images with a tool