LLM leaderboard: closed and open large language models
This LLM leaderboard follows Arena Text, where people blind-compare the answers of two language models and vote. The ranks below were taken on 16 October 2026, and next to each model stands the version from the board. Closed models work only through the maker’s service; open models can be downloaded and run on your own server.
In short
- General tasks, closed model: Gemini, rank 1 in Arena Text on 16 October 2026
- Texts, analysis and code, closed model: Claude, rank 3 in Arena Text on 16 October 2026
- Open weights, large model: Kimi, rank 18 in Arena Text on 16 October 2026
- Open weights, models of different sizes: Qwen, rank 24 in Arena Text on 16 October 2026
- Open model for your own server: GLM, rank 28 in Arena Text on 16 October 2026
- Cheap API with open weights: DeepSeek, rank 38 in Arena Text on 16 October 2026
The list
Order: Arena Text, checked 16 October 2026.
- #1
Gemini
Google, USA
- What it is good for
- General tasks, long context, video and documents, and links with Google services.
- Why it is on the list
- Place 1 on Arena Text as of 16 October 2026, model Gemini 4 Argon (High)
- 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
- #2
Claude
Anthropic, USA
- What it is good for
- Long documents, analysis, programming and agents that do tasks step by step.
- Why it is on the list
- Place 3 on Arena Text as of 16 October 2026, model Claude Opus 5.5 (High)
- 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
- #3
Kimi
Moonshot AI, China
- What it is good for
- A large open model with a 1 million token context for long documents and agents.
- Why it is on the list
- Place 18 on Arena Text as of 16 October 2026, model Kimi K3 (Max)
- Price
- see the plans on the site
- Access from Russia
- open model, downloadable with no country restrictions official page
- Official site
- kimi.com
- Glossary term
- Kimi
- #4
ChatGPT
OpenAI, USA
- What it is good for
- A general assistant and API: texts, spreadsheets, file analysis and voice.
- Why it is on the list
- Place 21 on Arena Text as of 16 October 2026, model gpt-5.6-sol-xhigh
- 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
- #5
Qwen
Alibaba (Qwen team), China
- What it is good for
- Open models from laptop to server, some under Apache 2.0.
- Why it is on the list
- Place 24 on Arena Text as of 16 October 2026, model qwen3.8-max
- 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
- #6
GLM (Z.ai)
Z.ai (Zhipu AI), China
- What it is good for
- Open models for programming and agents, smaller versions under MIT.
- Why it is on the list
- Place 28 on Arena Text as of 16 October 2026, model glm-5.3-max
- Price
- API: GLM-5.3-Flash from $0.15 per 1 million input tokens, GLM-5.3 from $1.4; chat price not checked (16 October 2026) plans
- Access from Russia
- open model, downloadable with no country restrictions official page
- Official site
- chat.z.ai
- Glossary term
- GLM (Z.ai)
- #7
Grok
SpaceXSI (SpaceXAI LLC, formerly xAI), USA
- What it is good for
- An assistant with a large context window and access to the X feed.
- Why it is on the list
- Place 37 on Arena Text as of 16 October 2026, model grok-4.20-beta1
- Price
- free plan; paid from $8 per month (The X Subscription Plus) (16 October 2026) plans
- Official site
- grok.com
- Glossary term
- Grok
- #8
DeepSeek
DeepSeek, China
- What it is good for
- Open models under MIT and a cheap API for large volumes of text.
- Why it is on the list
- Place 38 on Arena Text as of 16 October 2026, model deepseek-v4.1-flash-max
- 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
- #9
Gemma
Google DeepMind, USA
- What it is good for
- Compact open Google models under Apache 2.0 for your own server and laptop.
- Why it is on the list
- Place 76 on Arena Text as of 16 October 2026, model gemma-4-31b
- Price
- free (open weights) (16 October 2026) plans
- Access from Russia
- open model, downloadable with no country restrictions official page
- Official site
- ai.google.dev/gemma
- Glossary term
- Gemma
- #10
Vibe (formerly Le Chat)
Mistral AI, France
- What it is good for
- A European developer: the Vibe assistant and open models for part of the line.
- Why it is on the list
- Place 111 on Arena Text as of 16 October 2026, model mistral-large-4
- Price
- free plan; Pro $14.99 per month (16 October 2026) plans
- Official site
- chat.mistral.ai
- Glossary term
- Vibe (formerly Le Chat)
- #11
gpt-oss
OpenAI, USA
- What it is good for
- An open OpenAI model under Apache 2.0; the larger version fits on one graphics card.
- Why it is on the list
- Place 215 on Arena Text as of 16 October 2026, model gpt-oss-120b
- Price
- free (open weights) (16 October 2026) plans
- Access from Russia
- open model, downloadable with no country restrictions official page
- Official site
- openai.com/index/introducing-gpt-oss
- #12
Llama*
Meta, USA
- What it is good for
- Open models with a large ecosystem of fine-tuned versions.
- Why it is on the list
- Place 242 on Arena Text as of 16 October 2026, model llama-3.1-405b-instruct-bf16
- Price
- free (open weights, download on request) (16 October 2026) plans
- Russian language
- no per the site
- Official site
- dev.meta.ai/llama
- Glossary term
- Llama
In short: the top ranks on the check date
The six rows above come from Arena Text on 16 October 2026. The board has hundreds of models; we show the ranks of only those on this page and link to the board itself. On this date, Muse models from Meta stand above some of our candidates: they are not among this page’s candidates, so see their ranks on the board itself.
A rank in Arena shows whose answers people prefer more often in a blind comparison. It is not a score for your task: how to read such tables and where they go wrong is covered on the page about how rankings work.
Closed models
You cannot download a closed model: you send a request to the maker’s server and pay for the volume of text. On 16 October 2026, Gemini 4 Argon has rank 1 in Arena Text on 16 October 2026, Claude Opus 5.5 rank 3 in Arena Text on 16 October 2026, ChatGPT with the GPT-5.6 Sol model rank 21 in Arena Text on 16 October 2026, and Grok rank 37 in Arena Text on 16 October 2026.
The gap between neighboring ranks is often within the margin of error: Arena computes ranks with the Bradley-Terry model, and close models can share a row. So for work, choose from the top rows by price, language and access, not by a single rank.
Closed models get new features faster: search, files, voice and agents. You pay for this with dependence on the company, its prices and its list of countries.
If a company needs a closed model, check three things before paying: whether your country is on the official list, where requests are stored and whether they are used for training. The answers are in the service rules, not in the ranking.
Open models
Open models publish their weights: you can download them and run them on your own server. On 29 April 2025, the Qwen 3 family came out under Apache 2.0, and now Qwen has rank 24 in Arena Text on 16 October 2026. In August 2026, Alibaba for the first time opened the weights of a Qwen-Max class model with 2.4 trillion parameters.
Kimi has rank 18 in Arena Text on 16 October 2026: in June 2026, Moonshot AI released the weights of Kimi K3 with 2.8 trillion parameters. GLM has rank 28 in Arena Text on 16 October 2026, DeepSeek rank 38 in Arena Text on 16 October 2026, Gemma rank 76 in Arena Text on 16 October 2026, Mistral rank 111 in Arena Text on 16 October 2026, gpt-oss rank 215 in Arena Text on 16 October 2026, and Llama rank 242 in Arena Text on 16 October 2026.
The license matters more than the rank. Apache 2.0 and MIT allow commercial use with no revenue conditions, while the Qwen-Max, Kimi K3 and Llama licenses set conditions for large services. Each model’s license is on its card; more on the choice: open models for business. Which of them run on an ordinary computer, see the ranking of local neural networks.
Reasoning models
A reasoning model first writes a draft solution for itself and only then answers. This takes longer and costs more, but it is more accurate on math, code and multi-step tasks. In Arena Text, such modes show in the version name: High, Max, xhigh.
Open reasoning models appeared on 20 January 2025, when DeepSeek released R1 with weights under MIT. Now most models on this page have a reasoning mode.
For hard tasks judged by tests, not by votes, there is a separate page: reasoning models by the Artificial Analysis index. Definition of the term: reasoning model.
Price per million tokens
Through an API you pay for tokens: pieces of words at the input and output. On 16 October 2026, by the official pages, Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens. Gemini 4 Argon costs $2 and $10 until 30 November 2026, and $4 and $20 from 1 December.
At OpenAI, GPT-5.6 Sol costs $4 and $20, and GPT-6 Astra $10 and $50. Open models through the maker’s API are cheaper: DeepSeek-V4.1-Flash from $0.15 per million input tokens outside peak hours, GLM-5.3-Flash from $0.15 and GLM-5.3 from $1.4.
Count the price of a task, not the price of a token. A reasoning model spends several times more output tokens on its draft, so a model that is cheap by the price list can cost more per answer.
A cheap model is not always the better deal: if it makes twice as many mistakes, an employee spends time checking. A comparison of all API prices per task: API price comparison.
Which models are available in Russia
ChatGPT, Claude and Gemini do not officially support Russia: on 16 October 2026 it is not on the country lists of OpenAI, Anthropic and Google, while Kazakhstan is. For Grok, DeepSeek and Mistral there is no official data on Russia, and the terms need checking in their plans and service rules.
The open models Qwen, Kimi, GLM, gpt-oss and Gemma download with no country limits and run on your server. Then the data does not go abroad, and access does not depend on other people’s decisions.
We give no instructions on getting around restrictions. If a company needs a strong model inside its own network, the legal path is an open model on its own server.
How we built this top
- Rank in Arena Text on the check date: human votes in blind comparisons
- Open weights and license by the model card
- Price per million tokens by the official pages
- Access from Russia by the official country lists
Order: Arena Text. Places go only to tools whose models are on the board.
Checked:
How we build ratingsQuestions and answers
Which language model is first right now?
Can you run a strong LLM yourself?
Sources
- Arena, March 2026 update, March 2026arena.ai
- OpenAI Help Center, 16 October 2026help.openai.com
- Anthropic, 16 October 2026anthropic.com
- Google AI for Developers, 16 October 2026ai.google.dev
- Simon Willison, 20 January 2025simonwillison.net
- Simon Willison, 29 April 2025simonwillison.net
- Pravo.ru, ruling of the Tverskoy District Court of Moscow of 21 March 2022 (does not concern WhatsApp or users), 21 March 2022pravo.ru
- Anthropic, pricing, 16 October 2026claude.com
- Google AI for Developers, Gemini API pricing, 16 October 2026ai.google.dev
- OpenAI, API pricing, 16 October 2026developers.openai.com
- Arena Text, checked 16 October 2026arena.ai
* Meta is recognized as an extremist organization and its activity is banned in Russia.