Anthropic, Google, Meta, Apple, xAI, Mistral, DeepSeek, Qwen, open source and local models.
On one computer, the realistic options today are small models such as Gemma 4, Qwen3.8-27B, gpt-oss-20b or Nemotron 3.5 Lightning. Models above about 300 billion parameters need a server or a hosted service.
7 min read
Mistral published the Large 4 weights on October 29, 2026: about one trillion parameters, 49 billion active per token, trained in Europe.
5 min read
Gemini 4 Carbon is Google's strongest model, announced October 28, with a focus on cybersecurity and long engineering tasks.
4 min read
Gemini 4 Argon is a strong coding model with impressive results on Google's own large engineering projects, but on Google's public Android Bench it solves 21.3% of long tasks, behind the top Claude and GPT…
3 min read
Each lab picks its own tests, runs them at its own settings and inside its own coding agent, and the same model can get different scores from different publishers.
4 min read
On most shared coding tests Claude Opus 5.5 scores higher, including Google's independent Android Bench, where it solves 37.3% of long tasks against 29.3% for GPT-6.1 Sol.
4 min read
Six groups lead most of the field: DeepSeek, Alibaba's Qwen, Moonshot's Kimi, Z.ai's GLM, MiniMax and Xiaomi's MiMo.
5 min read
Moonshot AI is a Beijing company founded in March 2023 that makes the Kimi assistant and a line of very large open-weight models.
5 min read
Muse is Meta's family of AI models: hosted Muse Spark and Muse Wonder models that you use through Meta's Muse Code agent or API, plus two open-weight releases, the small Muse Glimmer and the very large Muse…
4 min read
Nemotron 3.5 is NVIDIA's family of open models for building agents, from a 4 GB edge model to a 120-billion-parameter vision model that runs on one DGX Spark or a single H200.
5 min read
Install a free runner such as Ollama or LM Studio, pick an open-weight model in a 4-bit quantized version whose file is clearly smaller than your graphics memory (or about half your RAM on an Apple Silicon…
5 min read
If you subscribe to Google AI Pro or Ultra, you can try Gemini 4 Argon now in the Gemini app and in AI Mode in Search; developers can call it through the paid tier of the Gemini API in Google AI Studio.
3 min read
Pick three or four candidate models, one per role you need (a strong model for hard work, a mid-priced one for daily work, a small one for volume), then run the same ten to twenty of your own tasks through…
4 min read
As of October 2026, the two strongest coding models that fit on one home machine are Alibaba's Qwen3.8-27B and Meta's Muse Glimmer-30B, both under the permissive Apache 2.0 licence.
3 min read
You can use Muse Wonder in two ways today, through Meta's Model API under the name muse-wonder-1.0 or inside Muse Code, Meta's coding agent.
4 min read
Grok 4.7 is a capable, low-priced coding model that SpaceXAI places close to the leaders on its own tests, but independent testing puts it a clear step behind Claude Opus 5.5, GPT-6.1 Sol and Gemini 4 Argon…
4 min read
Beam is the first open-weight model from Reflection, a US startup: 501 billion parameters, 23 billion active per token, text only, released on October 16, 2026, under Apache 2.0.
4 min read
Muse Spark Open is the downloadable version of Meta's Muse Spark, released on October 15, 2026, with 816 billion parameters, 42 billion active per token.
4 min read
Mistral Large 4 is available as a public preview through Mistral Studio and the Mistral API, where you select it under the name mistral-large-4.
4 min read
Muse Wonder and Muse Wonder Full are Meta's most capable AI models, released on October 12, 2026, and built for agents that carry out personal and professional tasks.
4 min read
A closed model stays on its maker's servers and you can only use it through an app or API under the company's terms; an open-weight model lets you download the trained weights and run, change or share them…
4 min read
Claude Haiku 5.5 is Anthropic's new small model, released on October 7, 2026, built for fast, repetitive, high-volume work such as summarizing, sorting and quick lookups.
5 min read
Kolibri-1 is an English-German open-weight model from Germany's Aleph Alpha, released on October 3, 2026, with 78 billion parameters and about 3.46 billion active per token.
4 min read
Claude Sonnet 5.5 is Anthropic's new mid-tier model, released on September 28, 2026, as a faster and lower-cost partner to Claude Opus 5.5.
5 min read
Choose GPT-6 Sol for demanding work such as coding, multi-step agent tasks and professional documents, and GPT-6 Luna for simple, repeated jobs at volume, such as sorting, extracting and summarizing.
4 min read
DeepSeek-V4.1-Flash, released on September 10, 2026, is an MIT-licensed model with 552 billion backbone parameters that activates only 8 billion per token when reading and 16 billion when writing.
5 min read
On August 28, 2026, Z.ai published GLM-5.3, a model of about 750 billion parameters, two weeks after its API launch, saying it needed extra safety testing because of the model's cyber abilities.
4 min read
Call the model deepseek-v4-pro through DeepSeek's API with the OpenAI or Anthropic format, give it the failing test and the code involved, and ask for a small, reviewable change.
4 min read
The simplest way is Qwen Studio, Alibaba's chat site, where Qwen3.8-Max can be picked as a model; developers can call it through Qwen Cloud with the OpenAI Python library.
4 min read
Install a recent build of llama.cpp (b10353 or newer), download Meta's 16.8 GB Q4_K_M file of Muse Glimmer from Hugging Face, and start llama-server with the --jinja option.
4 min read
Moonshot AI published the full Kimi K3 weights on Hugging Face on July 27, 2026: 2.8 trillion parameters in total, 104 billion active for each token, in a 1.56 TB download.
5 min read
Claude Opus 5 is Anthropic's new everyday flagship model, released on July 24, 2026, which the company says comes close to its top model, Claude Fable 5, at half the price.
5 min read
Grok 4.5 is SpaceXAI's new model for coding, agent tasks and office work, and today you reach it through Grok Build, Cursor or the SpaceXAI API under the name grok-4.5.
4 min read
Pick Terra as your everyday default, move up to Sol for long, specialized or high-stakes work, and drop down to Luna for short, repetitive jobs where speed and cost matter more than depth.
4 min read
Modern AI is trained and run on huge numbers of specialized chips in data centres, so whoever can build, buy and power that computing capacity can train bigger models, serve more users and keep control of the…
4 min read
Unless your organization is a vetted cyber-defense partner or joins one of Anthropic's trusted access programs, you cannot choose Mythos 5, so the real decision is between Claude Fable 5 and the cheaper…
4 min read
MiniMax M3 is an open-weight model of about 428 billion parameters, 23 billion of them active per token, with a one-million-token context window and native image and video input.
4 min read
Claude Fable 5 and Claude Mythos 5, launched by Anthropic on June 9, 2026, are the same new top-tier AI model offered in two ways.
5 min read
Claude Opus 4.8 is Anthropic's newest top model, released on May 28, 2026 at the same price as Opus 4.7, with better scores on coding and reasoning tests and a stronger habit of admitting uncertainty about…
5 min read
Call the model gemini-3.5-flash, lower its thinking level to low or minimal for simple jobs, and send work that can wait through the Batch API, which halves the price.
4 min read
Gemini 3.5 Flash is Google's newest AI model, launched on May 19, 2026 at Google I/O as the first of the Gemini 3.5 family, and it is now the default model in Google Search's AI Mode and in the Gemini app for…
5 min read
A fast model is the better choice when the task is well defined, the right answer is easy to check, and speed or cost matters, as in classification, extraction, reformatting and short replies.
4 min read
The quickest way is DeepSeek's own chat site or app, where the preview is available through Expert Mode and Instant Mode; developers can call deepseek-v4-pro or deepseek-v4-flash through DeepSeek's API using…
3 min read
Claude Design is a new Anthropic Labs tool, launched as a research preview on April 17, 2026, that lets you create designs, clickable prototypes, slide decks and one-pagers by describing them to Claude and…
4 min read
Claude Opus 4.7 is Anthropic's new most capable generally available model, released on April 16, 2026, with stronger results than Opus 4.6 on hard coding and long tasks, sharper vision and more literal…
4 min read
Muse Spark is a new AI model that Meta announced on April 8, 2026, the first from its Meta Superintelligence Labs team and the start of a new "Muse" series that will replace Llama inside Meta AI.
4 min read
Claude Mythos Preview is a new, unreleased Anthropic model that is so good at finding and exploiting software flaws that Anthropic is giving it only to a group of security partners, through an effort called…
4 min read
OpenClaw and Hermes Agent are free, open-source personal AI agents that you install yourself and talk to through apps like WhatsApp or Telegram; they can run commands, manage files and carry out tasks on your…
4 min read
Use GPT-5.4 Thinking for nearly all serious work, from reports and spreadsheets to coding and research, because it is fast enough, widely available and strong across OpenAI's published tests.
4 min read
Give Gemini 3.1 Pro a precise research question, ask it to plan before answering, to separate what sources say from its own reasoning, and to cite every factual claim.
3 min read
Install llama.cpp, then run "llama-cli -hf unsloth/Qwen3.5-35B-A3B-GGUF:Q4_K_M", which downloads a 21 GB compressed copy of Qwen3.5-35B-A3B and opens a chat on your own machine.
4 min read
Gemini 3.1 Pro, released by Google in preview on February 19, 2026, is an upgrade of Gemini 3 Pro focused on core reasoning, scoring a verified 77.1% on the ARC-AGI-2 puzzle test, more than double its…
4 min read
Claude Sonnet 4.6, released by Anthropic on February 17, 2026, is its new mid-sized model, now the default for Free and Pro users, with better coding, much stronger computer use and a 1-million-token context…
4 min read
Qwen3.5, released by Alibaba on February 16, 2026, is a new family of AI models whose first open-weight member, Qwen3.5-397B-A17B, reads text, images and video, supports 201 languages and is free to download…
4 min read
Claude Opus 4.6, released by Anthropic on February 5, 2026, is its most capable model, better at planning, long coding tasks, code review and everyday office work, and the first Opus model with a…
4 min read
On January 12, 2026, Apple and Google announced a multi-year collaboration under which the next generation of Apple's own AI models, the Apple Foundation Models, will start from Google's Gemini models and use…
4 min read
Cowork is a new mode in the Claude desktop app for Mac, released by Anthropic as a research preview on January 12, 2026, that lets Claude work directly in a folder you choose on your computer, reading,…
5 min read
Gemini 3 Flash is Google's new fast, low-cost model in the Gemini 3 family, released on December 17, 2025, which brings much of Gemini 3 Pro's reasoning to quick, everyday use.
5 min read
You can call Mistral Large 3 through a hosted API like any other model, or download its weights from Hugging Face and serve it yourself with vLLM on a single server with eight data-center GPUs.
4 min read
All three are strong enough for everyday professional work, and their published benchmark scores are close enough that the right choice depends on your tasks, your tools and your budget more than on a…
4 min read
Mistral 3 is a new family of open-weight AI models released by the French company Mistral AI on December 2, 2025, made of one large model, Mistral Large 3, and nine small Ministral 3 models in three sizes.
5 min read
DeepSeek-V3.2 is an open-weight AI model that the Chinese lab DeepSeek released on December 1, 2025, together with a high-reasoning variant called DeepSeek-V3.2-Speciale.
5 min read
Choose "Thinking" in the Gemini app's model menu to use Gemini 3 Pro, then give it material to work on, such as photos, long documents or a clear brief for a small app in Canvas, instead of short trivia…
4 min read
Claude Opus 4.5 is the model Anthropic released on November 24, 2025, as its most capable for coding, AI agents and computer use, and also for everyday work with documents, spreadsheets and slides.
5 min read
Nano Banana Pro, officially Gemini 3 Pro Image, is the image generation and editing model Google released on November 20, 2025, built on its Gemini 3 Pro language model.
5 min read
Gemini 3 Pro is the first model of Google's Gemini 3 generation, released in preview on November 18, 2025, and Google describes it as its most intelligent model for reasoning, understanding images and video,…
4 min read
Claude Haiku 4.5 is Anthropic's new small model, released on October 15, 2025, which Anthropic says codes about as well as Claude Sonnet 4 did five months ago, while costing a third as much and running over…
4 min read
Claude Sonnet 4.5 is Anthropic's new AI model, released on September 29, 2025, aimed at coding, computer use and long-running agents; Anthropic reports it can stay on a complex task for more than 30 hours and…
4 min read
Start with Free and upgrade only when you keep hitting a specific limit.
4 min read
Gemini 2.5 Flash Image is Google's new image generation and editing model, released on August 26, 2025, which had been tested anonymously under the nickname "nano-banana." It edits photos from plain-language…
5 min read
DeepSeek-V3.1, released on August 21, 2025, is one model that can either answer directly or think step by step first, replacing DeepSeek's separate chat and reasoning models with two modes of the same system.
5 min read
Leave GPT-5 on Auto for everyday requests, switch to Fast when you want a quick reply to a simple question, and pick Thinking when the task has several steps, a right answer you can check, or a cost if it…
4 min read
Claude Opus 4.1 is an upgrade to Anthropic's top model, Claude Opus 4, released on August 5, 2025, with better results on real-world coding, agentic tasks and detailed research, including 74.5% on the…
5 min read
Grok 4 is xAI's new flagship AI model, released on July 9, 2025, trained with large-scale reinforcement learning to reason and to use tools such as web search and a code interpreter on its own.
5 min read
Magistral is Mistral AI's first family of reasoning models, released on June 10, 2025, in two versions, an open 24-billion-parameter Magistral Small that anyone can download and a more capable Magistral…
4 min read
Claude Opus 4 and Claude Sonnet 4 are Anthropic's new AI models, released on May 22, 2025, built for coding, reasoning and long multi-step tasks; Opus 4 is the most powerful and can keep working on a task for…
5 min read
Choose the access route first and the model second. Mistral AI is a French company working under the GDPR, its API terms say customer data is not used for training, and its open-weight and enterprise models…
4 min read
You can try Qwen3 in the browser at chat.qwen.ai or download it to run locally with tools such as Ollama or LM Studio.
4 min read
Mistral Medium 3 is a mid-size language model that the French company Mistral AI released on May 7, 2025, for businesses.
5 min read
Qwen3 is a family of eight open-weight language models from Alibaba's Qwen team, released on April 28–29, 2025, ranging from a tiny 0.6-billion-parameter model to a 235-billion-parameter flagship.
5 min read
Only on unusually large home hardware: Llama 4 Scout has 109 billion parameters in total, so even a 4-bit version needs roughly 55 GB of memory for the weights alone, which points to a high-memory Mac or a…
4 min read
Llama 4 Scout and Llama 4 Maverick are Meta's newest open-weight AI models, released on April 5, 2025, and the first Llama models that understand images as well as text from the ground up and that use a…
5 min read
Upload the whole PDF, count its tokens to confirm it fits in the one-million-token window, and ask questions that require page numbers and short quotes for every claim.
3 min read
Gemini 2.5 Pro, released by Google on March 25, 2025, in an experimental version, is the first of a new Gemini generation in which every model reasons through a problem before answering.
4 min read
The model itself decides, token by token, when it has reasoned enough, so a hard question usually gets a longer hidden chain of thought than an easy one.
4 min read
Press Think when a question needs careful step-by-step work, such as math, logic, code or a plan with constraints, and press DeepSearch when the answer depends on current information from the web and X that…
4 min read
Claude 3.7 Sonnet, released by Anthropic on February 24, 2025, is a single model that can either reply right away or think through a problem step by step first, with that thinking shown to you.
5 min read
Grok 3 is the newest family of AI models from Elon Musk's company xAI, unveiled in a livestream on the evening of February 17, 2025, and trained with roughly ten times the computing power of Grok 2.
4 min read
Use GPT-4o for everyday writing, conversation, images, files and quick answers; use o3-mini when a math, science or coding problem needs careful step-by-step work and speed matters; use o1 for hard problems…
4 min read
Qwen2.5-Max is a large mixture-of-experts language model from Alibaba's Qwen team, released at the start of the Lunar New Year (January 28–29, 2025), trained on more than 20 trillion tokens and offered…
4 min read
You can try DeepSeek-R1 for free by turning on the DeepThink option at chat.deepseek.com or in the DeepSeek app, call it through the API as deepseek-reasoner, or download its open weights.
4 min read
On Monday, January 27, 2025, DeepSeek's free AI assistant became the most-downloaded free app in the US App Store, and worries that cheaper AI training could reduce demand for chips sent Nvidia's stock down…
4 min read
DeepSeek-R1 is a reasoning model released on 20 January 2025 by the Chinese lab DeepSeek, which reports performance on par with OpenAI's o1 in math, code and reasoning.
4 min read
DeepSeek-V3 is a large open-weights language model released on December 26, 2024, by the Chinese company DeepSeek, with 671 billion parameters of which only 37 billion work on each token.
5 min read
Sign in to Google AI Studio with a Google account, pick Gemini 2.0 Flash Experimental in the model menu, and start typing, uploading images or turning on tools such as Google Search.
4 min read
For most people, no. ChatGPT Plus at $20 a month already includes the full o1 model, and Pro's extra $180 only pays off if you hit usage limits often, rely on o1 pro mode for hard technical work, or make…
4 min read
Gemini 2.0 is Google's new generation of AI models, introduced on December 11, 2024, starting with an experimental Gemini 2.0 Flash that is faster than Gemini 1.5 Pro, can create images and speech as well as…
5 min read
On October 28, 2024, Apple released iOS 18.1, iPadOS 18.1 and macOS Sequoia 15.1, the free updates that bring the first Apple Intelligence features to recent iPhones, iPads and Macs.
5 min read
The easiest way is Ollama, a free app that downloads and runs a Qwen 2.5 model on your own computer with one command; the files and code examples are also on Hugging Face.
4 min read
Computer use is a new public beta feature, announced by Anthropic on October 22, 2024, that lets an upgraded Claude 3.5 Sonnet operate a computer the way a person does, by reading screenshots, moving the…
5 min read
Llama 3.2 is a set of openly downloadable AI models that Meta released on September 25, 2024, at its Connect conference.
5 min read
Use o1 when a task has one hard chain of logic to get right, such as a multi-step math problem, a tricky bug or a plan with many constraints.
4 min read
Grok is the chatbot from Elon Musk's company xAI, built into X: with a Premium or Premium+ subscription you open the Grok tab in an updated X app and type, choosing Grok-2 or the faster Grok-2 mini.
4 min read
Le Chat is Mistral AI's chat assistant at chat.mistral.ai: you sign in, choose a Mistral model such as the new Mistral Large 2, and chat as you would in ChatGPT, in English, French and many other languages.
4 min read
Llama 3.1 is a family of AI models Meta released on July 23, 2024, led by a 405-billion-parameter model that Meta says can compete with GPT-4o and Claude 3.5 Sonnet, plus upgraded 8B and 70B versions.
5 min read
Claude 3.5 Sonnet is Anthropic's newest AI model, released on June 20, 2024, which Anthropic says beats its previous top model, Claude 3 Opus, on most tests while running twice as fast and costing the same as…
5 min read
Apple Intelligence is Apple's new set of AI features for iPhone, iPad and Mac, announced at WWDC on June 10, 2024, that can rewrite and summarize text, create images and emoji, and give Siri a better grasp of…
5 min read
Start from the task, not the brand. Decide what matters most, such as quality, cost, privacy, how much text it must read at once, or which tools it needs, then shortlist two or three models and run the same…
4 min read
Copilot+ PCs are a new class of Windows laptops, announced by Microsoft on May 20, 2024, built with a dedicated AI chip called an NPU so that features like Recall, Cocreator and live caption translation run…
5 min read
Small language models are AI text models with a few billion parameters instead of hundreds of billions, light enough to run on a laptop and, in tests, on a phone.
4 min read
At its I/O conference on May 14, 2024, Google announced AI Overviews, AI-written summaries at the top of Search results, for everyone in the US; a faster, cheaper model called Gemini 1.5 Flash; a larger…
5 min read
Yes, all three allow commercial use, but on different terms: Mistral's open models use the permissive Apache 2.0 licence, while Meta's Llama 3 licence and Google's Gemma Terms of Use add conditions such as…
5 min read
Llama 3 is Meta's new family of downloadable language models, released on April 18, 2024, in two sizes, 8 billion and 70 billion parameters, trained on more than 15 trillion tokens of text.
5 min read
On March 17, 2024, xAI published the weights and architecture of Grok-1, the 314-billion-parameter language model behind its Grok chatbot, under the Apache 2.0 license, which allows free commercial use.
5 min read
The three Claude 3 models share the same basic abilities, including reading images and a 200,000-token context window, but trade intelligence against speed and price.
4 min read
Claude 3 is Anthropic's new family of AI models, announced on March 4, 2024, in three sizes: Haiku, the fastest and cheapest, Sonnet, the balanced middle option that now powers free claude.ai, and Opus, the…
5 min read
On February 26, 2024, the French company Mistral AI released Mistral Large, its most capable language model, which it ranks second only to GPT-4 among models available through an API, along with a smaller…
4 min read
A one-million-token context window lets a model read roughly 700,000 words, an hour of video or a large codebase in a single request, so you can ask questions across a whole body of material at once.
4 min read
Gemma is a family of lightweight open models that Google released on February 21, 2024, built from the same research and technology as its Gemini models.
4 min read
Both cost about $20 a month and give you each company's top model, Gemini Ultra 1.0 or GPT-4, but they bundle different things.
3 min read
Gemini 1.5 Pro is Google's new mid-size AI model, announced on February 15, 2024, that performs at about the level of its largest model, 1.0 Ultra, while using less computing power.
5 min read
On February 8, 2024, Google renamed its Bard chatbot to Gemini, after the family of AI models that powers it, and launched Gemini Advanced, a paid tier that runs on its largest model, Ultra 1.0.
4 min read
Since December 6, Google's free Bard runs on a tuned version of Gemini Pro, which Google says beats GPT-3.5, the model behind free ChatGPT, on six of eight standard benchmarks.
4 min read
Mixtral 8x7B is a language model released on December 11, 2023, by the Paris start-up Mistral AI, with open weights that anyone can download and use under the permissive Apache 2.0 license.
5 min read
Gemini is Google DeepMind's new family of AI models, announced on December 6, 2023, in three sizes: Ultra, Pro and Nano.
4 min read
Claude 2.1 is a new version of Anthropic's AI model, released on November 21, 2023, that can read up to 200,000 tokens at once, about 150,000 words or more than 500 pages, twice as much as Claude 2.0.
4 min read
Grok is a chatbot from Elon Musk's AI company xAI, unveiled on Saturday, November 4, 2023, and powered by a new language model called Grok-1.
5 min read
Choose a hosted model such as ChatGPT or Claude when you want the strongest answers with no setup, and an open-weight model such as Llama 2 or Mistral 7B when your data must stay on your own machines, you…
4 min read
Mistral 7B is the first language model from the French startup Mistral AI, released on September 27, 2023: a 7.3-billion-parameter model that anyone can download and use without restrictions under the Apache…
5 min read
Code Llama is a family of programming models that Meta released on August 24, 2023, built by training Llama 2 further on code, in sizes of 7, 13 and 34 billion parameters, free for research and commercial use.
4 min read
Qwen-7B is a 7-billion-parameter language model from Alibaba Cloud, released on August 3, 2023 together with a chat version, Qwen-7B-Chat, as downloadable weights that researchers and companies can use for…
4 min read
Llama 2 is a family of large language models released by Meta on July 18, 2023, in sizes of 7, 13 and 70 billion parameters, with chat-tuned versions, free for research and commercial use.
4 min read
Claude 2 is Anthropic's new AI model, released on July 11, 2023, with better scores in coding, math and reasoning than Claude 1.3, room for about 75,000 words in a single prompt, and longer answers.
4 min read
Anthropic has raised the amount of text Claude can take in at once from about 9,000 tokens to 100,000 tokens, roughly 75,000 words, or a whole short novel.
5 min read
PaLM 2 is Google's new large language model, announced at Google I/O on 10 May 2023, with stronger skills in more than 100 languages, in reasoning and math, and in writing code.
4 min read
Tongyi Qianwen is a large language model unveiled by Alibaba Cloud on 11 April 2023 that writes and converses in Chinese and English, and that Alibaba plans to build into all of its apps, starting with the…
4 min read
ChatGPT is a general-purpose assistant that writes from what it learned in training, Bing Chat searches the web first and links to its sources, and Google's Bard is a newer experiment that offers several…
3 min read
Bard is Google's conversational AI, built on a lightweight version of its LaMDA language model, and on 21 March 2023 Google began letting people in the US and the UK in through a waitlist at bard.google.com.
4 min read
Microsoft 365 Copilot is an AI assistant, announced on 16 March 2023, that will sit inside Word, Excel, PowerPoint, Outlook and Teams and draft, summarize and analyze using GPT-4 combined with your own work…
4 min read
Claude is a conversational AI assistant from Anthropic, a company founded by former OpenAI researchers, opened to businesses on March 14, 2023 through a chat interface and an API.
4 min read
LLaMA (Large Language Model Meta AI) is a family of four language models, from 7 billion to 65 billion parameters, that Meta released on February 24, 2023 for research use only.
4 min read
The new Bing is Microsoft's search engine rebuilt around a next-generation OpenAI language model, announced on February 7, 2023, that can chat with you, write complete answers from web results and cite its…
4 min read
Bard is an experimental conversational AI service that Google announced on February 6, 2023, built on its LaMDA language model and able to draw on information from the web for up-to-date answers.
4 min read
For everyday questions, drafting and back-and-forth conversation, the free ChatGPT preview is the easier choice.
4 min read
Constitutional AI is a training method that Anthropic described in a paper on 15 December 2022, in which an AI assistant learns to avoid harmful answers by critiquing and rewriting its own responses against a…
5 min read
Stable Diffusion 2.0, released by Stability AI on November 24, 2022, swaps in a new text encoder called OpenCLIP, trains on a dataset filtered to remove adult images, and adds a 768x768 model, a 4x upscaler,…
4 min read
GPT-3 is billed by the token, a piece of a word, and you pay for both the text you send and the text it writes back.
4 min read
Character.AI is a new website, opened to the public as a beta on 16 September 2022, where you can chat with AI characters and create your own by giving each one a name, a greeting and a description.
4 min read
A model card is a short document published with an AI model that explains what it is for, how it was trained, how it was tested and where it falls short.
4 min read
A language model is "open" to the degree that outsiders can get its weights (the learned numbers that make up the model), its code and the details of how it was trained, and use them under a license that…
4 min read
Imagen is a text-to-image system that Google Research unveiled on 23 May 2022, which turns a written description into a photorealistic picture by pairing a large language model with a chain of diffusion image…
5 min read
PaLM, short for Pathways Language Model, is a new Google language model with 540 billion parameters, trained on 6,144 of Google's TPU chips with a new system called Pathways.
5 min read
Ada, Babbage, Curie and Davinci are four GPT-3 models on OpenAI's API that trade speed and price against capability.
4 min read
LaMDA, short for Language Model for Dialogue Applications, is an experimental Google language model unveiled on 18 May 2021 at Google I/O that was trained on dialogue so it can hold open-ended conversations…
5 min read
Tokens are the small pieces of text a language model actually reads and writes: whole common words, parts of rarer words, punctuation marks and spaces.
5 min read
Google announced on 25 October 2019 that its search engine now uses BERT, a language model built on the Transformer design, to read each query as a whole sentence, so small words like "to," "for" and "no" are…
5 min read
Sign up for the newsletter and receive the free guide: what changed in AI, one lesson, one prompt to try. Unsubscribe in one click.