Claude is the best LLM for most businesses because Claude Opus 5.5 posts the highest Artificial Analysis Intelligence Index score and every current Claude model has a 1M-token context window. Gemini is the pick for multimodal work at a lower flagship price, GPT suits teams already on ChatGPT, and Kimi is the strongest choice for companies that want open weights they can host themselves.
Top 10 LLMs, compared| RANK | ENTRY | BEST FOR | SCORE | FROM |
|---|
| 1 |  Claude (Anthropic)Claude Opus 5.5 at max effort scored 58 on the Artificial Analysis Intelligence Index, the highest of any model on the leaderboard when read on 27 October 2026. | Agentic coding and long documents | 96.8 | Free plan; API from $0.10 per million input tokens |
|---|
|
| 2 |  Gemini (Google)Google reports that Gemini 4 Argon scores 68.9% on the Vals Index, against 67.0% for Claude Opus 5.5 and 63.1% for GPT-6 Astra. | Multimodal work at a lower flagship price | 95.7 | Free plan; API from $0.10 per million input tokens |
|---|
|
| 3 |  GPT (OpenAI)OpenAI lists GPT-6 Astra with a 1,050,000-token context window and a 128,000-token output limit. | Teams already standardised on ChatGPT | 94.9 | Free plan; API from $0.05 per million input tokens |
|---|
|
| 4 |  Muse (Meta)Meta says Muse Spark Open is the most capable US open-weight model to date. | Teams wanting a US open-weight option | 93.1 | Free (open weights); API from $1.25 per million input tokens |
|---|
|
| 5 |  Kimi (Moonshot AI)Moonshot says Kimi K3, at 2.8 trillion parameters, was the first open-source model to reach that size. | Self-hosting a frontier-scale model | 92.4 | Free (open weights); API $2 per million input tokens |
|---|
|
| 6 |  Grok (xAI)xAI's model documentation lists Grok 4.7 with a 500,000-token context window and text and image input. | Coding agents on a mid-range budget | 90.2 | Free plan; API from $1 per million input tokens |
|---|
|
| 7 |  DeepSeek (DeepSeek)DeepSeek-V4.1-Flash-1017, released on Hugging Face in October 2026, is a 763B-parameter model whose weights are MIT licensed. | High-volume work on a tight budget | 89.1 | Free (open weights); API from $0.15 per million input tokens |
|---|
|
| 8 |  Qwen (Alibaba)Alibaba Cloud's Singapore region gives new users 1M free tokens per Qwen model, valid for 90 days. | Cheap tokens with an open-weight choice | 88.3 | Free (open weights); API from $0.03 per million input tokens |
|---|
|
| 9 |  GLM (Z.ai)Z.ai says GLM-5.3 improves on GLM-5.2 by about 50% on its in-house Z.ai Code Bench. | Low-cost coding agents | 87.0 | Free (open weights); API from $0.07 per million input tokens |
|---|
|
| 10 |  Mistral (Mistral AI)Mistral says Mistral Large 4 has about 1 trillion total parameters, with 52 billion active. | EU data residency and open weights | 85.4 | Free plan; API from $0.10 per million input tokens |
|---|
|
|
How to choose
Start from what you need it for. Where each entry fits best:
Each entry's full verdict is below.
The ranking, in detail
01
Top-scoring models for coding and agents
Claude is Anthropic's model family: Claude Fable 5.1 for the hardest reasoning and long-running agent work, Claude Opus 5.5 for most workloads, and the faster Claude Sonnet 5.5 and Claude Haiku 5.5. All four have a 1M-token context window and run on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry.
Free plan; API from $0.10 per million input tokens
PROS- Highest Artificial Analysis index score
- 1M-token context on every model
CONS- Fable 5.1 costs $10 per million input
- No open-weight models
- Best for
- Agentic coding and long documents
- Where it falls short
- Fable 5.1 costs $10 per million input.
96.8
02
Google's multimodal models with a free tier
Gemini is Google's model family, led by Gemini 4 Argon, with Gemini 3.8 Flash and Flash-Lite models for faster, cheaper work across text, images, audio and video. Gemini 4 Argon ranked first on the Arena text leaderboard in its 20 October 2026 update, and the Gemini API has a free tier on its Flash models.
Free plan; API from $0.10 per million input tokens
PROS- First on the Arena text leaderboard
- Free API tier on Flash models
CONS- Argon price rises after launch offer
- No free API tier for Argon
- Best for
- Multimodal work at a lower flagship price
- Where it falls short
- Argon price rises after launch offer.
95.7
03
OpenAI's models behind ChatGPT
GPT is OpenAI's model family, led by GPT-6 Astra for demanding reasoning, coding and computer use, with GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna at lower prices. The models power ChatGPT and the OpenAI API, and GPT-6 Astra scored 53 on the Artificial Analysis Intelligence Index.
Free plan; API from $0.05 per million input tokens
PROS- Wide range, from Luna to Astra
- Largest listed context window here
CONS- Astra costs $10 per million input
- Free ChatGPT plan lacks GPT-6 Astra
- Best for
- Teams already standardised on ChatGPT
- Where it falls short
- Astra costs $10 per million input.
94.9
04
Meta's frontier models, with open weights
Muse is Meta's model family, which has replaced Llama on Meta's developer site; Muse Wonder, released on 12 October 2026, targets coding, long-running agents and computer use with a 1,048,576-token context window. Muse Spark 1.3 is the cheaper API model, Muse Spark Open is the open-weight release, and the subscriber-only Muse Wonder Full scored 56 on the Artificial Analysis Intelligence Index.
Free (open weights); API from $1.25 per million input tokens
PROS- 1M-token context window
- Open-weight Muse Spark Open available
CONS- Muse Wonder still in preview
- Wonder Full limited to Muse Code
- Best for
- Teams wanting a US open-weight option
- Where it falls short
- Muse Wonder still in preview.
93.1
05
Open-weight models at frontier scale
Kimi is Moonshot AI's model family, led by Kimi K3.1 Pro, which Moonshot calls its most capable 3T-class open-weight model, with a 1M-token context window and strengths in coding and agent tasks. K3.1 Pro scored 55 on the Artificial Analysis Intelligence Index, and Moonshot says its full weights will be released by 4 November 2026.
Free (open weights); API $2 per million input tokens
PROS- Top open-weight benchmark score
- 1M-token context window
CONS- K3.1 Pro weights not yet released
- Self-hosting needs very large hardware
- Best for
- Self-hosting a frontier-scale model
- Where it falls short
- K3.1 Pro weights not yet released.
92.4
06
xAI's models for coding and agents
Grok is xAI's model family (xAI now brands itself SpaceXSI on its own sites), led by Grok 4.7 for coding, agentic tasks and knowledge work. Grok 4.7 scored 46 on the Artificial Analysis Intelligence Index and costs $2 per million input tokens for prompts under 200,000 tokens.
Free plan; API from $1 per million input tokens
PROS- 500,000-token context window
- Low output price for a flagship
CONS- Lower index score than top five
- Higher input price over 200K tokens
- Best for
- Coding agents on a mid-range budget
- Where it falls short
- Lower index score than top five.
90.2
07
Low-cost open-weight models
DeepSeek is the Chinese lab behind the DeepSeek V4 family, led by DeepSeek V4 Pro and DeepSeek V4.1 Flash, both with 1M-token context windows and strengths in reasoning and coding. DeepSeek publishes the weights on Hugging Face and charges half price on its API outside peak hours.
Free (open weights); API from $0.15 per million input tokens
PROS- MIT-licensed open weights
- Half-price API off peak
CONS- Peak-hour API prices double
- Large models need multi-GPU servers
- Best for
- High-volume work on a tight budget
- Where it falls short
- Peak-hour API prices double.
89.1
08
Alibaba's open and hosted model family
Qwen is Alibaba's model family, led by Qwen3.8 Max for reasoning and coding with a 1M-token context, plus Qwen3.7 Plus and Qwen3.8 Flash as cheaper tiers. Alibaba also publishes open-weight Qwen3.8 models on Hugging Face, including Qwen3.8-27B, and Qwen3.8 Max scored 45 on the Artificial Analysis Intelligence Index.
Free (open weights); API from $0.03 per million input tokens
PROS- Very low entry API price
- Open-weight Qwen3.8 models
CONS- Free quota limited to Singapore
- High cost per task on Max
- Best for
- Cheap tokens with an open-weight choice
- Where it falls short
- Free quota limited to Singapore.
88.3
09
Open-weight models for coding agents
GLM is the model family from Z.ai, led by GLM-5.3 for complex coding and long-running agent tasks, with GLM-5.3 Flash as a cheaper tier. GLM-5.3 is a 753B-parameter model with weights on Hugging Face, and it scored 45 on the Artificial Analysis Intelligence Index.
Free (open weights); API from $0.07 per million input tokens
PROS- Free API models available
- Open weights on Hugging Face
CONS- Custom licence on GLM-5.3
- Lower index score than top five
- Best for
- Low-cost coding agents
- Where it falls short
- Custom licence on GLM-5.3.
87.0
10
European models with open weights
Mistral AI is the French lab behind Mistral Large 4, released in preview on 6 October 2026, alongside Mistral Medium 3.5, Mistral Small 4 and the small Ministral 3 models. Mistral Large 4 is a multimodal model that Mistral says will ship with open weights, served from its European data centres.
Free plan; API from $0.10 per million input tokens
PROS- Served from European data centres
- Apache 2.0 licence on small models
CONS- Lowest index score on this list
- Large 4 still in preview
- Best for
- EU data residency and open weights
- Where it falls short
- Lowest index score on this list.
85.4
Put your company on this ranking
Ask us for a quote for one, two or three years. Nothing recurring.
Positions #4 to #10 are available to larger companies. Contact us.
Not on this ranking?Tell us about your company and we will reply by email.
Frequently asked questions
How much does an LLM API cost?
LLM APIs charge per million tokens, and prices range widely: Qwen starts at $0.03 per million input tokens and GPT at $0.05, while flagship models such as GPT-6 Astra and Claude Fable 5.1 cost $10 per million input tokens. Most vendors, including Claude, Gemini and Mistral, also offer a free chat plan, and discounts such as DeepSeek's off-peak pricing cut costs further.
How do I choose an LLM for my business?
Start with the job: coding agents and long documents favour Claude or GPT, multimodal work suits Gemini, and high-volume tasks suit cheaper models such as DeepSeek or Qwen. Then check how each vendor handles your data, whether the model runs on your cloud provider, and whether you need open weights to host it yourself, as Kimi, GLM and Mistral allow.
What is the best LLM for a small business?
Gemini and Claude both have free plans and small models priced at $0.10 per million input tokens, which suits small teams testing AI on a budget. Mistral is a sensible choice for European firms that want EU hosting, while GPT through ChatGPT is a familiar start for staff who already use it.
Should I use an open-weight or a closed model?
Closed models such as Claude, Gemini and GPT currently post the highest benchmark scores and need no infrastructure to run. Open-weight families such as Kimi, DeepSeek, Qwen and GLM let you host the model yourself for data control and lower unit costs, though the largest ones need multi-GPU servers.
Which LLM has the largest context window?
GPT-6 Astra lists the largest window at 1,050,000 tokens, just ahead of the roughly 1M-token windows on Claude, Muse, Kimi, DeepSeek and Qwen. Grok 4.7 offers 500,000 tokens, which is still enough for most long contracts or large codebases.
Some links on Lister are affiliate links. Rankings are not affected.