New AI tools in 2026: releases by month
New AI tools in 2026 appear almost every week, but few change a company’s choice. This page lists only releases with a date and a maker’s announcement: Claude Opus 5.5, Gemini 4 Argon, DeepSeek-V4, Qwen3.8, Gemma 4, Mistral Large 4, Manus 2.0, and the robots Atlas and GR00T. We update the timeline once a month; the last check was on 16 October 2026.
In short
- Anthropic models: Claude Opus 5.5, Sonnet 5.5, and Haiku 5.5, September–October
- Google’s flagship: Gemini 4 Argon, September–October
- DeepSeek open models: V4 in April, V4.1-Flash in September
- Alibaba’s open flagship: Qwen3.8, August
- Google’s open models: Gemma 4, April
- A European model: Mistral Large 4, October
- An agent: Manus 2.0, September
- Robots: the production Atlas in January, the GR00T N1.7 model in March
The list
The order follows the criteria of this page; it is not a rating.
- #1
Claude
Anthropic, USA
- What it is good for
- Anthropic’s assistant; the Claude 5.5 model family came out in September–October 2026.
- Why it is on the list
- On 28 May 2026, Anthropic raised $65 billion at a $965 billion valuation; the company’s annualized revenue passed $47 billion. Anthropic, 28 May 2026
- Price
- free plan; paid from $17 per month billed annually ($20 billed monthly) (16 October 2026) plans
- Russian language
- partly per the site
- Access from Russia
- not officially supported official page
- Official site
- claude.ai
- Glossary term
- Claude
- #2
Gemini
Google, USA
- What it is good for
- Google’s assistant; the Gemini 4 Argon model came out in fall 2026.
- Why it is on the list
- On 11 August 2026, Sundar Pichai said that the Gemini app has more than 1 billion monthly active users. TechCrunch, 11 August 2026
- Price
- free plan; paid from $4.99 per month (Google AI Plus), Google AI Pro $19.99 per month (16 October 2026) plans
- Russian language
- yes per the site
- Access from Russia
- not officially supported official page
- Official site
- gemini.google.com
- Glossary term
- Gemini
- #3
DeepSeek
DeepSeek, China
- What it is good for
- Open models under MIT; DeepSeek-V4 and V4.1-Flash came out in 2026.
- Why it is on the list
- On 10 September 2026, DeepSeek released the open model DeepSeek-V4.1-Flash with 552 billion parameters and cut its API prices. DeepSeek, 10 September 2026
- Price
- chat and app free; API from $0.15 per 1 million input tokens (DeepSeek-V4.1-Flash, off-peak hours) (16 October 2026) plans
- Official site
- chat.deepseek.com
- Glossary term
- DeepSeek
- #4
Qwen
Alibaba (Qwen team), China
- What it is good for
- Alibaba’s open models; the Qwen3.8 flagship was opened in August 2026.
- Why it is on the list
- In August 2026, Alibaba opened the weights of a Qwen-Max class model for the first time: Qwen3.8 with 2.4 trillion parameters, 95 billion of them active. Qwen, model card on Hugging Face, 8 August 2026
- Price
- see the plans on the site
- Access from Russia
- open model, downloadable with no country restrictions official page
- Official site
- chat.qwen.ai
- Glossary term
- Qwen
- #5
ChatGPT
OpenAI, USA
- What it is good for
- OpenAI’s assistant: 900 million weekly users as of February 2026; officially not available in Russia.
- Why it is on the list
- On 27 February 2026, OpenAI said that ChatGPT has 900 million weekly active users. TechCrunch, 27 February 2026
- Price
- free plan; paid from $8 per month (Go), Plus $20 per month (16 October 2026) plans
- Russian language
- yes per the site
- Access from Russia
- not officially supported official page
- Official site
- chatgpt.com
- Glossary term
- ChatGPT
- #6
Veo
Google, USA
- What it is good for
- Video from a description by Google; Veo 3.1 Lite is available in the Gemini API.
- Why it is on the list
- On 15 October 2025, Google released Veo 3.1 and said that by then more than 275 million videos had been created on Veo models in the Flow service. Google, The Keyword, 15 October 2025
- Price
- no free plan; in the Gemini API from $0.05 per second of video (Veo 3.1 Lite, 720p); in the Gemini app it is included in paid Google AI subscriptions (16 October 2026) plans
- Russian language
- yes per the site
- Access from Russia
- not officially supported official page
- Official site
- deepmind.google/models/veo
- Glossary term
- Veo
- #7
Isaac GR00T
NVIDIA, USA
- What it is good for
- An open NVIDIA model for humanoid robots, with a right to commercial use.
- Why it is on the list
- On 16 March 2026, NVIDIA presented Isaac GR00T N1.7, an open model for humanoid robots licensed for commercial use; Humanoid, LG Electronics, NEURA and Noble Machines announced that they use it. NVIDIA Newsroom, 16 March 2026
- Price
- free: open weights and code (16 October 2026) plans
- Official site
- developer.nvidia.com/isaac/gr00t
- #8
Atlas
Boston Dynamics (Hyundai), USA
- What it is good for
- A production electric humanoid from Boston Dynamics for factories and warehouses.
- Why it is on the list
- On 5 January 2026, Boston Dynamics showed the production electric Atlas: 56 degrees of freedom, lifts up to 50 kg and swaps its own batteries; all 2026 deliveries are allocated between Hyundai and Google DeepMind. Boston Dynamics, 5 January 2026
- Status
- Since 21 September 2026, Atlas has been trained on logistics and parts sorting tasks at the RMAC test center at the Hyundai Metaplant America site near Savannah (USA); these are trials, and there is no data on regular work on the production line. source
- Price
- see the plans on the site
- Russian language
- no per the site
- Official site
- bostondynamics.com/atlas
In short: the 10 main releases of the year
The list includes releases that change a company’s choice: new flagships from Anthropic and Google, open modelsopen weights model. An open weights model is a neural network whose parameter files you can download and run on your own hardware. you can run on your own server, agents, and robots. To the lines above we add a tenth release: on 31 August, Google released TimesFM-3 for forecasting.
Within each quarter, releases go by date. We do not include news without a maker’s announcement or a date, even if every aggregator writes about it. The overall map of the section with all rankings by task is in the Top AI review.
January–March
On 5 January, Boston Dynamics showed the production electric Atlas: 56 degrees of freedom, lifting up to 50 kg, and the robot changes its own batteries. All deliveries for 2026 are split between a Hyundai plant and Google DeepMind.
Atlas status as of fall: since 21 September 2026, the robot has been trained on logistics and parts sorting tasks at a Hyundai test center near Savannah. These are trials; there is no data on regular work on an assembly line.
On 27 February, OpenAI reported that ChatGPT has 900 million weekly active users. This is not a new model but a mark of scale from which to estimate demand for AI assistantsAI assistant. An AI assistant is a helper built on a neural network that, at a person’s request, answers questions, prepares texts and does simple tasks in work programs. among your employees and customers.
On 16 March, NVIDIA presented the open Isaac GR00T N1.7 model for humanoids with a right to commercial use. Humanoid, LG Electronics, NEURA, and Noble Machines announced they will use it.
April–June
On 2 April, Google released Gemma 4 under Apache 2.0. The smaller E2B and E4B run on a phone with no internet, and the larger 26B and 31B on a personal computer. For a company, this is an open model you can build into your product with no royalties.
On 24 April, DeepSeek opened a preview of DeepSeek-V4: V4-Pro with 1.6 trillion parameters, 49 billion of them active, and V4-Flash with 284 billion. A context of 1 million tokens became standard in all the company’s services.
Both releases are open: the weights are on Hugging Face, and you can run the model on your own server in Russia without depending on the maker. Which models fit your hardware is covered in the top local AI modelslocal LLM. A local LLM is a neural network that runs on your own computer or server, not in a vendor’s cloud, so your data goes nowhere..
July–September
In August, Alibaba for the first time opened the weights of a Qwen-Max class model: Qwen3.8 with 2.4 trillion parameters, 95 billion of them active. The flagship comes under Alibaba’s own Qwen license, with conditions for services with more than 100 million users a month or revenue over $50 million, while Qwen3.8–27B from the same family is under Apache 2.0 and fits on one workstation.
On 13 August, DeepSeek made V4-Pro generally available, and on 10 September it released DeepSeek-V4.1-Flash with 552 billion parameters and cut API prices.
On 31 August, Google released TimesFM-3 for time series forecasting: 330 million parameters, trained on a trillion data points. The weights are for non-commercial use only and the code is under Apache 2.0, so these weights cannot be used in a commercial sales forecast.
On 22 September, Anthropic released Claude Opus 5.5: according to the company, it is 40% cheaper than Opus 5 on typical tasks, at $4 per 1 million input tokenstoken. A token is a piece of text that a language model works with: a word, part of a word, or a symbol. and $20 per 1 million output tokens. On 28 September, Claude Sonnet 5.5 and Manus 2.0 came out: the new Manus agent used 23.2% fewer tokens in a test setup.
October–December
On 30 September, Google announced Gemini 4 Argon and at first opened it only to cybersecurity partners, Reuters reported. In October, the model appeared in the Gemini API: according to Google’s documentation, the total context limitcontext window. The context window is the amount of text, counted in tokens, that a language model sees at one time: instructions, documents, the question and its own answer. is 1,048,576 tokens. Google does not include Russia in the list of Gemini API regions.
On 6 October, Mistral AI presented Mistral Large 4 with 1 trillion parameters, trained on 3,800 GPUsGPU. A GPU (graphics processing unit) is a graphics processor: a chip that performs thousands of identical calculations at the same time, so neural networks are trained and run on it. in its own data centers in Europe. The company promises to publish the weights by the end of October.
On 7 October, Anthropic released Claude Haiku 5.5, the smallest model of the family: $0.10 per 1 million input tokens for requests up to 100,000 tokens.
Tesla promises to launch the first Optimus line by the end of 2026, with a future capacity of 1 million robots a year. This is a plan: the company has not published delivery data.
We will add November and December as releases come out. The timeline includes only what the maker announced itself, with a date.
How not to miss what matters
Follow the makers’ news pages: at Anthropic, Google, and DeepSeek they carry dates and model names. Aggregators retell rumors, and some of those releases are never confirmed.
Watch the open rankings. Arena compares models through blind pairwise votes by users, separately for text, code, images, video, and agents. If a new model has not appeared in the ranking, there are no quality scores yet.
For a company, what matters is not every release but the one that changes the price or quality of your process. We track such changes in our AI market monitoring.
How we built this top
- The maker has an announcement with a date
- The release changes a company’s choice: price, open weights, new tasks
- The model or product is available, not only promised
The order follows the criteria of this page; it is not a rating.
Checked:
How we build ratingsQuestions and answers
Which AI tools came out in 2026?
Sources
- Arena, March 2026 update, March 2026arena.ai
- Boston Dynamics, 5 January 2026bostondynamics.com
- TechRepublic, 12 October 2026techrepublic.com
- DataNorth, 1 September 2026datanorth.ai
- Google, company blog, 2 April 2026blog.google
- DeepSeek, API news, 24 April 2026api-docs.deepseek.com
- Anthropic, 22 September 2026anthropic.com
- Anthropic, 7 October 2026anthropic.com
- Manus, 28 September 2026manus.im
- Reuters via Khaleej Times, 1 October 2026khaleejtimes.com
- Google AI for Developers, 14 October 2026ai.google.dev
- Mistral AI, 6 October 2026mistral.ai