OpenAI is testing Dots, its always-on personal agents, with a small number of people on the $20 a month ChatGPT Plus plan. The agents are also opening to every ChatGPT Business Standard user. Dots can work without a person watching, connect to a computer, and reach more than 4,000 apps. They began on GPT-6 Astra and can now run on GPT-6.1 Terra as well.
Google introduced Gemini 4 Neon for coding, knowledge work, and science, between its Carbon and Hydrogen models. It costs $3 per million input tokens, $0.15 for cached input, and $15 per million output tokens, with a 1 million token window. Google says the highest setting scores 61 on the Artificial Analysis Intelligence Index, third overall, and 70.9 percent on the Vals Index, second place. It is in the Gemini API and in the Gemini app for Pro and Ultra subscribers.
GoogleQuick look +
Google introduced Gemini 4 Neon for coding, knowledge work, and science, between its Carbon and Hydrogen models. It costs $3 per million input tokens, $0.15 for cached input, and $15 per million output tokens, with a 1 million token window. Google says the highest setting scores 61 on the Artificial Analysis Intelligence Index, third overall, and 70.9 percent on the Vals Index, second place. It is in the Gemini API and in the Gemini app for Pro and Ultra subscribers.
A browser game packs a galaxy of 150 million star systems into a 299 kilobyte HTML file. You fly a ship through it, and the whole thing is small enough to load as one page.
3D, VR & spatialQuick look +
A browser game packs a galaxy of 150 million star systems into a 299 kilobyte HTML file. You fly a ship through it, and the whole thing is small enough to load as one page.
The scale is the trick. A galaxy that size usually needs a huge install. This one is smaller than most photos.
OpenAI is testing Dots with a small number of ChatGPT Plus users and opening them to every Business Standard user. What to check before an always-on agent gets account access.
Our analysis · written by handQuick look +
When should you try ChatGPT Dots, and what should you check before you let one into your accounts?
OpenAI is testing Dots with a small number of ChatGPT Plus users and opening them to every Business Standard user. What to check before an always-on agent gets account access.
OpenAI released GPT-6.2 Astra for harder projects and GPT-6.2 Sol for everyday work, in the API as gpt-6.2-astra and gpt-6.2-sol. Astra costs $8 per million input tokens and $40 per million output tokens, about 20 percent below GPT-6 Astra, and cached input is $0.40 per million. Sol stays at $2 per million input tokens and $10 per million output tokens. Both are in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users.
OpenAIQuick look +
OpenAI released GPT-6.2 Astra for harder projects and GPT-6.2 Sol for everyday work, in the API as gpt-6.2-astra and gpt-6.2-sol. Astra costs $8 per million input tokens and $40 per million output tokens, about 20 percent below GPT-6 Astra, and cached input is $0.40 per million. Sol stays at $2 per million input tokens and $10 per million output tokens. Both are in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users.
This race track, the forest around it and the beat-up 1996 Fiat Uno with a ladder on the roof are all built from code. No textures, no 3D models, no physics library, just WebGPU.
3D, VR & spatialQuick look +
This race track, the forest around it and the beat-up 1996 Fiat Uno with a ladder on the roof are all built from code. No textures, no 3D models, no physics library, just WebGPU.
The builder gave Fable 5.2 one detailed prompt plus a critic subagent that checked its work against photos of the real car. It runs free in your browser.
Tesla invited selected customers to order a Cybercab for personal use at $29,990, plus a required $299 a month Autonomy Subscription and a $1,000 non-refundable deposit. The car is limited to cities where Tesla already runs its robotaxi service, has no steering wheel, and the company says canceling the subscription leaves it unusable on public roads. Order terms bar commercial use, including adding the car to Tesla's own robotaxi network, and a federal safety audit of the steering-wheel-free design is still open.
The VergeQuick look +
Tesla invited selected customers to order a Cybercab for personal use at $29,990, plus a required $299 a month Autonomy Subscription and a $1,000 non-refundable deposit. The car is limited to cities where Tesla already runs its robotaxi service, has no steering wheel, and the company says canceling the subscription leaves it unusable on public roads. Order terms bar commercial use, including adding the car to Tesla's own robotaxi network, and a federal safety audit of the steering-wheel-free design is still open.
Press B in this Minecraft mod and the world locks into LEGO bricks you can knock down and rebuild. It plays like the old LEGO games, with fixed cameras, TNT walls, and a level editor.
GamesQuick look +
Press B in this Minecraft mod and the world locks into LEGO bricks you can knock down and rebuild. It plays like the old LEGO games, with fixed cameras, TNT walls, and a level editor.
The mod is called Brickify. The full build video is on the creator's channel, and the download is free.
The King's Trust and OpenAI launched a partnership on November 16, backed by a £1 million donation from OpenAI. The program is meant to help up to 25,000 young people in the UK build practical AI skills for work over 18 months, with priority for those not in education, employment or training. It sits inside a wider set of nationally recognized AI qualifications the charity plans to launch.
OpenAIQuick look +
The King's Trust and OpenAI launched a partnership on November 16, backed by a £1 million donation from OpenAI. The program is meant to help up to 25,000 young people in the UK build practical AI skills for work over 18 months, with priority for those not in education, employment or training. It sits inside a wider set of nationally recognized AI qualifications the charity plans to launch.
Physicist Nathaniel Craig described how GPT-6 Astra helped develop a candidate lattice definition of the Standard Model. The November 15 post explains work presented in an October preprint, including assistance with the construction and detailed proof. It is a proposed mathematical approach under scientific scrutiny, rather than an established resolution of the problem.
OpenAIQuick look +
Physicist Nathaniel Craig described how GPT-6 Astra helped develop a candidate lattice definition of the Standard Model. The November 15 post explains work presented in an October preprint, including assistance with the construction and detailed proof. It is a proposed mathematical approach under scientific scrutiny, rather than an established resolution of the problem.
OpenAI introduced ChatGPT Wallet for a limited rollout to ChatGPT Pro users in the US through dots. Users review each purchase and approve it with a security check; agents receive payment tokens instead of actual card details. The service was built with American Express, Mastercard and Visa and preserves existing card rewards and eligible dispute protections.
OpenAIQuick look +
OpenAI introduced ChatGPT Wallet for a limited rollout to ChatGPT Pro users in the US through dots. Users review each purchase and approve it with a security check; agents receive payment tokens instead of actual card details. The service was built with American Express, Mastercard and Visa and preserves existing card rewards and eligible dispute protections.
Google's model for complex work you run often, between Carbon and Hydrogen
NewAdded Nov 18, 2026
Gemini 4 Neon is Google's November 18, 2026 model for coding, knowledge work, and science. The listed API price is $3 per million input tokens, $0.15 per million cached input tokens, and $15 per million output tokens, with a 1 million token window. Google says it is in the Gemini API, in Gemini for business, and for Pro and Ultra subscribers in Google Antigravity and the Gemini app.
Why: Google places Neon between Carbon, for the hardest and longest projects, and Hydrogen, for fast everyday work. The price sits under the other current frontier list prices, which makes it the one to try when a hard job still has to run often. Google's index figures are Google's report, not a CuratedAI test.
OpenAI's November 17 frontier model, about 20 percent below GPT-6 Astra on the list price
NewAdded Nov 17, 2026
GPT-6.2 Astra is OpenAI's November 17, 2026 model for harder projects, released with GPT-6.2 Sol. The API name is gpt-6.2-astra. OpenAI's listed API price is $8 per million input tokens, $0.40 per million cached input tokens, and $40 per million output tokens. OpenAI says it is in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users.
Why: It is the current OpenAI frontier model a team can call, at a lower list price than GPT-6 Astra. Use it when a missed detail is expensive. A lab rank is not a reason to move a whole queue onto it.
The cheaper GPT-6.2, still at $2 per million input tokens
NewAdded Nov 17, 2026
GPT-6.2 Sol is the lower-priced model OpenAI released with GPT-6.2 Astra on November 17, 2026. The API name is gpt-6.2-sol. The listed API price stays at $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens, the same standard rates as GPT-6.1 Sol. OpenAI says it is in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users.
Why: The list price did not go up, so this is the GPT-6.2 to try first on work you can check. Keep Astra for the cases where a miss is expensive. An unchanged price is not proof the model behaves like GPT-6.1 Sol.
Anthropic's November 2026 Fable model: better coding, reasoning and vision, up to 30% faster, and cheaper
New this monthAdded Nov 10, 2026
Claude Fable 5.2 is Anthropic's November 9, 2026 update to Fable 5.1, released alongside Claude Mythos 5.2, which is the same underlying model with fewer cyber and biology safeguards. Anthropic says it improves on coding, reasoning and vision, generates output up to 30% faster than Fable 5.1, and uses fewer tokens per task at Medium effort. On the Claude Platform it is claude-fable-5-2 at $8 per million input tokens and $40 output, down from $10 input; a fast mode that generates up to 2.5 times faster costs $16 and $80. It is available on the Claude Platform, AWS, Google Cloud and Microsoft Azure, and in Claude plans that include Fable.
Why: It is the strongest generally available Claude model, and the price went down rather than up. Anthropic's own launch table puts it ahead of Opus 5.5, GPT-6 Astra and Gemini 4 Carbon on most of the agentic coding and computer-use benchmarks it reports, and it blocks far fewer legitimate requests than Fable 5.1. Anthropic still recommends Opus 5.5 as the everyday model for most coding and knowledge work, with Fable 5.2 as the advisor or orchestrator above it.
The restricted-access twin of Fable 5.2, for vetted cybersecurity and life-sciences teams
New this monthAdded Nov 10, 2026
Claude Mythos 5.2 is the same underlying model as Claude Fable 5.2, released November 9, 2026, without some of Fable's cybersecurity and biology safeguards. Access is restricted: security professionals apply through Anthropic's Cyber Verification Program, life-sciences researchers through the Life Sciences Verification Program, and Claude Enterprise customers get it inside Claude Security. Anthropic reports it finds more vulnerabilities than Opus 5.5 and Mythos 5.1, and it shares Fable 5.2's listed price of $8 per million input tokens and $40 output.
Why: For the teams who can get it, it is the strongest Claude for vulnerability research and protein design, and Anthropic reports it is much less likely than Mythos 5.1 to reach outside its sandbox in adversarial tests. It ranks low here only because most readers cannot use it.
Meta's smallest, cheapest Muse model, built to do the repetitive work inside agents
New this monthAdded Nov 10, 2026
Muse Ember is Meta Superintelligence Labs' November 9, 2026 model, the fourth under the Muse brand and its lightest. Meta built it for the work agents repeat all day: tool calls, coding subtasks, routing, extraction and long-document parsing, and pitches it as the worker paired with Muse Wonder as the planner. It takes text, images and audio, so an agent can read screenshots, charts and scanned documents or work from voice memos and recordings without a second model. TechCrunch reports it at 10 cents per million input tokens.
Why: Cheap worker models are where agent bills are won or lost, and Ember arrives priced at 10 cents per million input tokens with multimodal input built in. Meta's launch chart puts it ahead of Anthropic's Haiku and OpenAI's Luna on cost per task against the Artificial Analysis index, though that comparison is Meta's own.
Thinking Machines’ second-generation model for visual coding and customization
New this monthAdded Nov 7, 2026
Released November 5, 2026, Inkling 2 Small accepts text and images and produces text. Thinking Machines reports a 594B-parameter mixture of experts with 17B active per token and adjustable reasoning effort. Inference and fine-tuning are available through Tinker.
Why: A new option to evaluate for visual development and custom-model workflows. Its launch availability should not be confused with the planned weight release.
Generate and revise video clips up to 15 seconds long
New this monthAdded Nov 7, 2026
Google announced Gemini Omni 1.2 Flash on November 5, 2026, with general availability through the Gemini API and Gemini for business. It generates and edits clips from 3 to 15 seconds, with reference inputs and conversational revisions. Google lists 720p generation at $0.07 per second.
Why: Worth evaluating when a short clip needs several controlled revisions rather than a complete restart.
Restricted-access model for authorized vulnerability research
New this monthAdded Nov 7, 2026
GPT-6 Cyber became available November 5, 2026 for approved Daybreak Red organizations. OpenAI documents text and image input, text output, a 1,050,000-token context window and up to 128,000 output tokens. Standard short-context API pricing is $20 per million input tokens and $100 per million output tokens.
Why: An important specialist release to track separately from general coding models because access is restricted.
Image generation and editing for detailed visual work
New this monthAdded Nov 7, 2026
Released November 4, 2026, GPT Image 2.6 Sunburst accepts text and images for generation and editing. OpenAI positions it for detailed visual work. Its documentation points to complexity controls, markup-guided editing, SVG and animated GIF output. Image output is priced at $30 per million tokens; input charges are separate.
Why: A distinct named API model worth evaluating against your existing image workflow.
Released November 4, 2026, GPT Image 2.6 Flare accepts text and images for generation and editing. OpenAI positions it for fast iteration. Its documentation points to complexity controls, markup-guided editing, SVG and animated GIF output. Image output is priced at $30 per million tokens; input charges are separate.
Why: A distinct named API model worth evaluating against your existing image workflow.
Moonshot's new 3-trillion-class flagship, with a 1M context and weights promised by November 4
New this monthAdded Oct 23, 2026
Kimi K3.1 Pro is the first model in Moonshot's K3.1 series, released October 23, 2026. Moonshot calls it the most capable model in its 3-trillion-parameter class, with a 1 million token context. Reasoning is always on, at low, high or max effort, and the default is max. The API model id is kimi-k3.1-pro. A Highspeed variant, which Moonshot says writes about two to three times faster, is sold separately. Moonshot says the full weights will be published by November 4. On Moonshot's own Terminal-Bench 4.0 chart, K3.1 Pro scores 65.7 percent, against 58.2 percent for GPT-6 Astra and 66.4 percent for Claude Opus 5.5.
Why: K3 is no longer the Kimi the independent index measures. K3.1 Pro is the model Moonshot is selling as the successor, already on the API, with open weights still a dated promise rather than a download.