Best LLMs in 2026: 20 Large Language Models Ranked

The best large language models in 2026 come from Google, Anthropic and OpenAI, in that order on the most-cited independent benchmark. Meta's new Muse Wonder is fourth, and the best open model, Moonshot AI's Kimi K3.1 Pro, is fifth. Below are the 20 LLMs that matter now, ranked with public leaderboards, plus their API prices, context windows and licenses, and the best pick for coding, writing, low cost, local use and Europe.

Everything was checked on November 8, 2026. The frontier moves every few weeks: Anthropic, OpenAI and Google all shipped new flagship models between September and October.

Top 10 LLMs in 2026, ranked

One model per maker, ranked by the Artificial Analysis Intelligence Index and checked against Arena's text leaderboard:

1. Gemini 4 (Google): the top score on both leaderboards

2. Claude Opus 5.5 (Anthropic): the best LLM for coding

3. GPT-6 Astra (OpenAI): OpenAI's most capable model, at a premium price

4. Muse Wonder (Meta): Meta's first frontier model, #3 on Arena

5. Kimi K3.1 Pro (Moonshot AI): the best open-weights model

6. Grok 4.7 (SpaceXAI, formerly xAI): a 500,000-token context at $2/$6

7. Hy4 (Tencent): Tencent's new flagship, formerly Hunyuan

8. Qwen3.8-Max (Alibaba): Alibaba's best, with open base weights

9. DeepSeek V4.1 Flash (DeepSeek): the most-used model on OpenRouter, MIT-licensed

10. GLM-5.3 (Z.ai): a strong open model with a cheap Flash version

Key numbers

• Google leads, narrowly. Gemini 4 Carbon scores 60.0 on the Artificial Analysis Intelligence Index, just ahead of Claude Opus 5.5 (59.1) and GPT-6 Astra (56.9).

• Open models trail by about 13 points. The best open-weights model, Kimi K3.1 Pro, scores 46.5. Nine of the ten highest-scoring open models are Chinese; the tenth is France's Mistral Large 4.

• Claude leads coding. Claude Opus 5.5 is first on Arena's WebDev leaderboard (1,814), ahead of Claude Sonnet 5.5 and GPT-6 Astra.

• A million tokens is standard. Every current flagship from Anthropic, OpenAI and Google reads about one million tokens of context.

• Prices split. Anthropic and OpenAI cut prices this autumn, while Google is doubling Gemini 4 Argon's price on December 1, 2026.

LLM leaderboards

Two leaderboards are cited more than any others. Artificial Analysis runs the same set of hard tests on every model (reasoning, coding, maths, agent tasks) and combines them into one Intelligence Index; version 5 scores top out around 60. Arena (formerly LMArena) shows voters two anonymous answers to the same prompt and turns millions of votes into an Elo-style score. They measure different things, so a model can rank higher on one than the other: OpenAI's models do much better on Artificial Analysis than in Arena's text votes.

Arena text leaderboard, top 10 from the board dated November 6, 2026 (9.1 million votes, 419 models):

Artificial Analysis Intelligence Index, each model's best setting, checked November 8, 2026:

The 20 best LLMs compared

API prices are per million input and output tokens, from each maker's pricing page on November 8, 2026, unless marked as Artificial Analysis's figure (AA). The index is the Artificial Analysis Intelligence Index.

The top 10 LLMs in detail

1. Gemini 4 (Google)

Google's Gemini 4 family tops both leaderboards. Gemini 4 Argon, generally available since October 13, 2026, is the version you can actually use: #2 on Arena's text board and #1 on its creative writing board. The stronger Gemini 4 Carbon (October 28) scores 60.0 on Artificial Analysis, the highest of any model, but Google has opened it only to cyber defenders and selected partners so far.

• Price: Argon $2/$10 per million tokens until November 30, then $4/$20 (Google AI Studio)

• Also: Gemini 3.8 Flash ($0.75/$3.75 until December 31) is #9 on Arena, and Gemma 4 is Google's open family under Apache 2.0

2. Claude Opus 5.5 (Anthropic)

Claude Opus 5.5 (September 22, 2026) is a point behind Gemini 4 Carbon on Artificial Analysis and first on Arena's WebDev coding board, which makes it the best LLM for coding in most comparisons. It reads a million tokens at the standard price, with no long-context surcharge. Its cheaper sibling Claude Sonnet 5.5 ($2/$10) scores 70.6% on Terminal-Bench 4.0 in Anthropic's own test, above Opus, and Claude Haiku 5.5 ($0.10/$0.50) is the strongest of the cheap models.

• Price: $4/$20 per million tokens

• Also: Claude Fable 5.1 ($10/$50) is the same model as Mythos 5.1, which Anthropic keeps for vetted cyber and life-science users

3. GPT-6 Astra (OpenAI)

GPT-6 Astra (September 2026) is OpenAI's most capable model: third on Artificial Analysis and third on Arena's WebDev board, and the first OpenAI model rated "Critical" for cyber capability under its Preparedness Framework. It does much worse in Arena's text votes (#36), where voters prefer Gemini, Claude and Muse answers. GPT-6.1 Sol (54.6 on the index) is close behind at a fifth of the price, and GPT-6.1 Terra (51.3) costs $1/$5.

• Price: $10/$50 per million tokens; Sol $2/$10; Terra $1/$5; Luna $0.10/$0.50 (OpenAI pricing)

• Context: 1.05 million tokens, up to 128,000 out

4. Muse Wonder (Meta)

Meta Superintelligence Labs released Muse Wonder, its first frontier-class model, on October 12, 2026. It is #3 on Arena's text board. The stronger Muse Wonder Full (50.1 on Artificial Analysis) is only available inside Meta's Muse Code subscription. Meta moved away from open Llama models in April with the closed Muse Spark, then partly returned to open weights in October with Muse Spark Open.

• Price: $4/$20 per million tokens on the Meta Model API (TechCrunch)

• Llama: no new Llama weights since Llama 4 in April 2025

5. Kimi K3.1 Pro (Moonshot AI)

Kimi K3.1 Pro from Beijing's Moonshot AI is the best open-weights LLM: 46.5 on Artificial Analysis, and the only open model in the top 10 of Arena's WebDev board. It went on Moonshot's API on October 24, 2026, with weights following in early November. In Artificial Analysis's own Terminal-Bench 4.0 run it scored 67.2%, ahead of Gemini 4 Carbon (65.2%) and Claude Sonnet 5.5 (63.6%).

• Price: about $2/$10 per million tokens (Artificial Analysis)

• License: a modified MIT license with a revenue clause for companies that sell it as a service

6. Grok 4.7 (SpaceXAI)

Grok 4.7 (September 2026) is the most capable model from xAI, which SpaceX bought in February 2026 and renamed SpaceXAI in July. It scores 35.5 on Artificial Analysis, well behind the top four, but its price and 500,000-token context window make it a reasonable value pick. It powers the Grok assistant.

• Price: $2/$6 per million tokens below 200,000 tokens; $4/$12 above (xAI docs)

7. Hy4 (Tencent)

Tencent renamed its Hunyuan models "Hy". Hy4, a mixture-of-experts model with 770 billion parameters (49 billion active per token), came out as an open preview in August 2026 and scores 34.6 on Artificial Analysis. Its predecessor Hy3 is open under Apache 2.0.

• Price: about $0.83/$2.50 per million tokens (Artificial Analysis)

8. Qwen3.8-Max (Alibaba)

Qwen3.8-Max (August 2026) is Alibaba's flagship: 33.7 on Artificial Analysis and #23 on Arena's text board. The API model is closed, but Alibaba released the 2.4-trillion-parameter base weights under a custom license, and its smaller Qwen3.8-27B is Apache 2.0, one of the best models to run on your own GPU.

• Price: about $2/$6 per million tokens (Artificial Analysis)

9. DeepSeek V4.1 Flash (DeepSeek)

DeepSeek's V4.1 Flash (updated October 17, 2026) is open under the MIT license and the most-used model on OpenRouter by tokens. It scores 32.6 on Artificial Analysis with 552 billion parameters, only 16 billion active per token, which keeps it fast and cheap. DeepSeek charges half price off-peak.

• Price: $0.30/$1.20 per million tokens; $0.15/$0.60 off-peak

• Context: 1 million tokens

10. GLM-5.3 (Z.ai)

Z.ai (formerly Zhipu) released GLM-5.3 in August 2026: 31.7 on Artificial Analysis and #26 on Arena's text board, with 753 billion parameters. Its smaller GLM-5.3-Flash is MIT-licensed and among the cheapest capable models anywhere.

• Price: $1.40/$4.40 per million tokens; Flash $0.15/$0.50 (Z.ai pricing)

• License: GLM-5.3 uses a modified MIT license with a clause for hosting businesses

10 more LLMs worth knowing

• Inkling 2 Small (Thinking Machines Lab, November 5, 2026): 30.6 on Artificial Analysis with only 17 billion active parameters; Apache 2.0 weights are promised (Thinking Machines).

• MiMo-V2.6-Pro (Xiaomi, September 2026): MIT-licensed, 29.9 on Artificial Analysis, and strong at long-context reasoning.

• MiniMax-M3.1-Flash (MiniMax, October 2026): open weights, 29.1, at about $0.15/$0.60.

• Mistral Large 4 (Mistral AI, October 2026): Europe's best open model, 26.1, under a license that lets companies self-host it but requires a commercial license to resell it as a service.

• Reflection Beam 1.0 (Reflection AI, October 2026): an Apache 2.0 open model built to rival the Chinese open models (TechCrunch).

• Nemotron 3.5 Super VL (NVIDIA, October 2026): open, 120 billion parameters, and NVIDIA's most capable model on the index.

• Command A+ (Cohere, May 2026): built for enterprise retrieval and agents, with a 128,000-token context (Cohere).

• MAI-Thinking-1 (Microsoft AI, August 2026): Microsoft's own reasoning model, in public preview on Azure Foundry (Microsoft AI).

• Mercury 2.5 (Inception, September 2026): a diffusion LLM that writes whole blocks at once, at about 1,100 tokens per second.

• Apertus v1.5 (Swiss AI, July 2026): fully open, including its training data, under Apache 2.0 (Hugging Face).

Best LLM by use case

• Coding: Claude Opus 5.5, first on Arena's WebDev board (1,814). For value, Claude Sonnet 5.5 ($2/$10) is second (1,772). The best open model for code is Kimi K3.1 Pro. See our best AI coding agents for the tools built on these models.

• Writing and chat: Gemini 4 Argon leads Arena's creative writing board (1525), ahead of Claude Opus 5.5 (1518).

• Low cost: Claude Haiku 5.5 and GPT-6 Luna both cost $0.10/$0.50, and Haiku scores 36.5 on the index, above many full-size models. DeepSeek V4.1 Flash is the cheapest strong open model.

• Long documents: Claude reads a million tokens at the standard price; most other providers charge more above 200,000–272,000 tokens.

• Running locally: Qwen3.8-27B (Apache 2.0) runs on a single 24 GB GPU at 4-bit, and Google's Gemma 4 (Apache 2.0) comes in sizes from 2 billion to 31 billion parameters. OpenAI's gpt-oss-20b is another Apache 2.0 option.

• European data residency: Mistral runs its models in the EU through regional endpoints and lets you self-host Mistral Large 4 (Mistral). Anthropic's API offers only global or US processing; EU routes go through AWS Bedrock or Google Cloud.

What changed in 2026

• LMArena became Arena in January (Arena), and SpaceX bought xAI in February.

• Meta went closed, then partly open again: the closed Muse Spark replaced Llama in April; Muse Spark Open followed in October.

• The best open models are Chinese: Kimi, DeepSeek, GLM, MiMo and MiniMax hold the top open spots, and Chinese open models made up about 61% of the tokens on OpenRouter by May 2026 (Wing VC).

• The strongest models are gated: Anthropic's Mythos 5.1 and Google's Gemini 4 Carbon are restricted to vetted users because of their cyber capabilities.

• Capability got cheaper fast: the price of a fixed level of performance has fallen about 47% per quarter since 2023 (Epoch AI). Claude Opus 5.5 costs $4/$20, against $15/$75 for Opus 4.1.

• Companies spend most on Anthropic: Anthropic holds 55% of enterprise LLM API spending, against 31% for OpenAI and 9% for Google, in a Menlo Ventures survey of about 400 US companies (Menlo Ventures, November 2026; Menlo is an Anthropic investor).

FAQ

What is the best LLM in 2026?

Google's Gemini 4 and Anthropic's Claude Opus 5.5 are effectively tied at the top: Gemini 4 Carbon scores 60.0 on the Artificial Analysis Intelligence Index and Opus 5.5 scores 59.1. Gemini leads Arena's text votes; Claude leads coding. OpenAI's GPT-6 Astra is third.

What is the best LLM for coding?

Claude Opus 5.5, which is first on Arena's WebDev leaderboard, followed by Claude Sonnet 5.5 and GPT-6 Astra. Kimi K3.1 Pro is the best open model for coding.

What is the best open-source LLM?

Kimi K3.1 Pro has the highest score of any open-weights model (46.5 on Artificial Analysis), under a modified MIT license. For a plain permissive license, DeepSeek V4.1 Flash (MIT) and Qwen3.8-27B (Apache 2.0) are the strongest.

Is ChatGPT an LLM?

ChatGPT is an app built on OpenAI's LLMs, currently the GPT-6 family. Gemini, Claude, Grok, Meta AI and Kimi are likewise apps built on each company's own models.

Which LLM has the largest context window?

The current flagships from OpenAI (1.05 million tokens), Anthropic and Google (1 million) are about level. Subquadratic's SubQ claims 12 million tokens but is in private beta.

What is the cheapest good LLM API?

Claude Haiku 5.5 and GPT-6 Luna, at $0.10 per million input tokens and $0.50 per million output tokens. Among open models, DeepSeek V4.1 Flash and GLM-5.3-Flash cost a few cents to a few tenths of a dollar per million tokens.

How are LLMs ranked?

Benchmarks like the Artificial Analysis Intelligence Index run the same tests on every model. Arena collects votes on anonymous head-to-head answers. They often disagree, so this list uses both.

How this list was built

The list: the companies tagged "Large Language Model" in LeadWave's database, plus the makers that rank high but weren't tagged yet: Anthropic, Tencent and Xiaomi. We left out companies that compress or fine-tune other companies' models rather than training their own. One model per maker, the strongest one.

The ranking: the Artificial Analysis Intelligence Index (checked November 8, 2026), cross-checked with Arena's text and WebDev leaderboards (November 6, 2026).

The facts: release dates, prices, context windows and licenses checked on November 8, 2026 against each maker's own announcements, pricing pages, model cards and Hugging Face pages, or Artificial Analysis where a maker's page wasn't available (marked AA).

Looking for the companies rather than the models? See our list of AI labs.

Selling to AI companies?

LeadWave's list of AI companies covers 9,000 AI companies with CEO details. See also our list of AI labs, the best AI coding agents and the best AI image generators.

About this list

Compiled by LeadWave from the public LLM leaderboards and company announcements, checked November 8, 2026. First published in January 2025 as a list of 12 models and fully rewritten in November 2026.