Which LLM is ahead in 2026?
In the Vals Index, Claude Sonnet 5.5 leads at 70.85%, ahead of Claude Opus 5.5 at 70.18%.
Maybe you aren't hunting for the top scorer anyway. You want to know how many language models exist, what they cost, and how much context they handle.
In this article, you'll find the key LLM statistics for 2026 on providers, prices, parameters, and context windows.
Start with the short version:
- Our model catalogue contains 298 records across 36 developer groups. Its newest saved release date is October 26, 2026. The catalogue includes Gemini 4 Argon, which became generally available on October 13.
- In the Vals Index dated October 27, 2026, Claude Sonnet 5.5 scores 70.85%, Claude Opus 5.5 70.18%, and Gemini 4 Argon 69.24%. Vals Index v2.1 weights tasks in finance, coding, legal work, and tax by their share of the U.S. economy.
- Selected API base prices are $0.1 / $0.5 for GPT-6 Luna, $0.75 / $3.75 for Gemini 3.8 Flash, and $4 / $20 for Claude Opus 5.5 per 1 million input and output tokens. Context tiers, introductory offers, and regions change the effective price.
- Providers publish total and active parameter counts for some open models, including Kimi K3 at 2.8 trillion total and 104 billion. Many hosted models have no official parameter count.
- Ten million tokens do not always mean a 10-million-token context window. Meta says Llama 4 Scout supports that context length; Qwen-Long processes 10-million-token documents through file upload. Gemini 4 Argon has a shared context window of 1,048,576 tokens across input, thinking, and output. Synchronous responses allow up to 65,536 output tokens; Long Decode can use more of the remaining context budget.
1. How Many Large Language Models Are There in 2026?
This catalogue contains 298 records across 36 developer groups. Gemini 4 Argon was announced on September 30 and became generally available on October 13. Qwen 3.8 Flash Next is listed separately in the open-weight directory.
According to the 2026 Stanford AI Index, industry produced more than 90% of notable frontier models in 2025.
2. The Biggest LLM Providers by Model Count
Who leads depends on the yardstick. Here, the number of models in the catalogue sits next to each provider's share of token volume at businesses.
2.1 Catalogue Models by Developer
The chart shows the catalogue models by developer:
OpenAI has the largest record set at 43, followed by Alibaba at 43 and Google at 29.
2.2 Token Volume Share at Ramp Businesses
The chart shows each provider's share of token volume among businesses that pay through Ramp:
Anthropic and OpenAI together account for about 97% of token volume in this slice.
3. Parameters and Architecture: What Providers Disclose
Many providers do not publish parameter counts for their hosted frontier models. For open weights, some providers publish both total and per-token active parameters.
3.1 Total Size of Open Models
Moonshot lists Kimi K3 at 2.8 trillion total and 104 billion, plus 16 selected experts per token. DeepSeek-V3 is listed with 671 billion core parameters and 37 billion; its downloadable weights also contain a separate MTP module that is not part of this core count.
Most examples in the chart use Mixture-of-Experts architectures. Their active parameter count is smaller than the total size. Our catalogue classifies 111 records as MoE.
Mistral Large 4 is included at the centrally recorded total size. Its launch post reports a rounded 1 trillion total parameters and the versioned model page a more precise 1.05 trillion. Both report 52 billion active parameters.
The interactive table uses the central parameter dataset. Rows marked "estimated" are based on analyses or leaks, not provider disclosures.
Mistral's Small 4 launch post lists 6 billion active parameters per token, or 8 billion including embedding and output layers. Its Hugging Face card says 6.5 billion activated per token, and the table uses that figure. For most current frontier models, the parameter column deliberately reads "unknown":
Legend:
Showing 298 models
Model | Developer | Parameters |
|---|---|---|
MiniMax M2.7 MoE | MiniMax | Unknown |
MiniMax M2.5 MoE | MiniMax | Unknown |
GLM-4.7 MoE | Z.ai | Unknown |
GPT-6.1 Sol | OpenAI | Unknown |
GPT-6.1 Terra | OpenAI | Unknown |
GPT-6 Astra | OpenAI | Unknown |
GPT-6 Sol | OpenAI | Unknown |
GPT-6 Luna | OpenAI | Unknown |
GPT-5.6 Sol | OpenAI | Unknown |
GPT-5.6 Terra | OpenAI | Unknown |
GPT-5.6 Luna | OpenAI | Unknown |
GPT-5.5 | OpenAI | Unknown |
GPT-5.5 Pro | OpenAI | Unknown |
GPT-5.5 Instant | OpenAI | Unknown |
ChatGPT chat-latest | OpenAI | Unknown |
GPT-5.4 | OpenAI | Unknown |
GPT-5.4 Pro | OpenAI | Unknown |
GPT-5.4 mini | OpenAI | Unknown |
GPT-5.4 nano | OpenAI | Unknown |
GPT-5.3-Codex | OpenAI | Unknown |
GPT-5.3 Instant | OpenAI | Unknown |
GPT-5.2 | OpenAI | Unknown |
GPT-5.1 Instant | OpenAI | Unknown |
GPT-5.1 Thinking | OpenAI | Unknown |
GPT-5 | OpenAI | Unknown |
GPT-5 Pro | OpenAI | Unknown |
GPT-5 mini | OpenAI | Unknown |
GPT-5 nano | OpenAI | Unknown |
GPT-4 Turbo | OpenAI | Unknown |
GPT-4.1 | OpenAI | Unknown |
GPT-4.1 mini | OpenAI | Unknown |
GPT-4.1 nano | OpenAI | Unknown |
GPT-3.5 Turbo | OpenAI | Unknown |
o3 | OpenAI | Unknown |
o3-pro | OpenAI | Unknown |
o3-mini | OpenAI | Unknown |
o4-mini | OpenAI | Unknown |
o1 | OpenAI | Unknown |
o1-mini | OpenAI | Unknown |
Claude Opus 5.5 | Anthropic | Unknown |
Claude Sonnet 5.5 | Anthropic | Unknown |
Claude Haiku 5.5 | Anthropic | Unknown |
Claude Fable 5.1 | Anthropic | Unknown |
Claude Mythos 5.1 | Anthropic | Unknown |
Claude Fable 5 | Anthropic | Unknown |
Claude Mythos 5 | Anthropic | Unknown |
Claude Sonnet 5 | Anthropic | Unknown |
Claude Opus 5 | Anthropic | Unknown |
Claude Opus 4.8 | Anthropic | Unknown |
Claude Opus 4.7 | Anthropic | Unknown |
Claude Opus 4.6 | Anthropic | Unknown |
Claude Sonnet 4.6 | Anthropic | Unknown |
Claude Opus 4.5 | Anthropic | Unknown |
Claude Opus 4.1 | Anthropic | Unknown |
Claude Sonnet 4.5 | Anthropic | Unknown |
Claude Haiku 4.5 | Anthropic | Unknown |
Claude Sonnet 4 | Anthropic | Unknown |
Claude Opus 4 | Anthropic | Unknown |
Claude 3.7 Sonnet | Anthropic | Unknown |
Claude 3.5 Haiku | Anthropic | Unknown |
Gemini 4 Carbon | Google DeepMind | Unknown |
Gemini 4 Argon | Google DeepMind | Unknown |
Gemini 3.8 Flash | Unknown | |
Gemini 3.8 Flash Cyber | Unknown | |
Gemini 3.7 Flash | Unknown | |
Gemini 3.6 Flash | Unknown | |
Gemini 3.5 Flash | Unknown | |
Gemini 3.1 Pro Preview | Unknown | |
Gemini 3 Flash Preview MoE | Unknown | |
Gemini 3.5 Flash-Lite | Unknown | |
Gemini 3.1 Flash-Lite MoE | Unknown | |
Gemini 2.5 Pro MoE | Unknown | |
Gemini 2.5 Flash MoE | Unknown | |
Gemini 2.5 Flash-Lite MoE | Unknown | |
Gemini 3 Pro MoE | Unknown | |
Gemini 2.0 Flash MoE | Unknown | |
Gemini 1.5 Pro MoE | Unknown | |
Grok 4.7 | xAI | Unknown |
Grok 4.6 | xAI | Unknown |
Grok 4.5 | xAI | Unknown |
Grok 4.3 | xAI | Unknown |
Grok 4.20 Reasoning | xAI | Unknown |
Grok 4.20 Multi-Agent | xAI | Unknown |
Grok Build 0.1 | xAI | Unknown |
Grok 4 | xAI | Unknown |
Grok 3 | xAI | Unknown |
Grok 2 | xAI | Unknown |
Qwen 3.7 Max MoE | Alibaba | Unknown |
Seed 1.8 | ByteDance | Unknown |
Seed 2.0 Pro | ByteDance | Unknown |
Muse Spark 1.3 | Meta | Unknown |
Muse Wonder | Meta | Unknown |
Muse Wonder Full | Meta | Unknown |
Muse Spark 1.2 | Meta | Unknown |
KAT-Coder-Air V2.5 | Kuaishou | Unknown |
KAT-Coder-Pro V2.5 | Kuaishou | Unknown |
Seed 1.6 Flash | ByteDance | Unknown |
Seed 2.0 Lite | ByteDance | Unknown |
Seed 2.0 Mini | ByteDance | Unknown |
Seed 2.0 Code | ByteDance | Unknown |
Seed 2.1 Turbo | ByteDance | Unknown |
Qwen 3.7 Flash | Alibaba | Unknown |
Qwen 3.7 Plus MoE | Alibaba | Unknown |
Amazon Nova 2 Lite | Amazon | Unknown |
Amazon Nova Premier | Amazon | Unknown |
Amazon Nova Pro | Amazon | Unknown |
Amazon Nova Lite | Amazon | Unknown |
Amazon Nova Micro | Amazon | Unknown |
Sonar | Perplexity | Unknown |
Sonar Pro | Perplexity | Unknown |
Sonar Reasoning Pro | Perplexity | Unknown |
Sonar Deep Research | Perplexity | Unknown |
Solar Pro 4 | Upstage | Unknown |
Kimi K3.1 Pro | Moonshot AI | 2.8T* |
Kimi K3 MoE(104B active) | Moonshot AI | 2.8T |
Qwen 3.8 2.4T A95B MoE(95B active) | Alibaba | 2.4T |
Qwen 3.8 Max 0902 MoE | Alibaba | 2.4T |
Claude 3 Opus | Anthropic | 2T* |
Llama 4 Behemoth MoE(288B active) | Meta | 2T |
GPT-4 MoE(220B active) | OpenAI | 1.76T* |
DeepSeek-V4-Pro MoE(49B active) | DeepSeek | 1.6T |
LongCat 2.0 MoE(48B active) | Meituan | 1.6T |
Mistral Large 4 MoE(52B active) | Mistral AI | 1.05T |
Ring 2.6 1T MoE(63B active) | inclusionAI | 1T |
Kimi K2.5 MoE(32B active) | Moonshot AI | 1T |
Kimi K2 Thinking MoE(32B active) | Moonshot AI | 1T |
Kimi K2 0711 MoE(32B active) | Moonshot AI | 1T |
Kimi K2 0905 MoE(32B active) | Moonshot AI | 1T |
Kimi K2.6 MoE(32B active) | Moonshot AI | 1T |
Kimi K2.7 Code MoE(32B active) | Moonshot AI | 1T |
Qwen 3.6 Max-Preview MoE | Alibaba | 1T* |
Yi-Large MoE | 01.AI | 1T |
MiMo-V2.5-Pro MoE(42B active) | Xiaomi | 1T |
MiMo-V2.5-Pro-UltraSpeed MoE(42B active) | Xiaomi | 1T |
Inkling MoE(41B active) | Thinking Machines Lab | 975B |
GLM-5.2 MoE(40B active) | Z.ai | 744B |
GLM-5.3 MoE(40B active) | Z.ai | 744B |
GLM-5.1 MoE(40B active) | Z.ai | 744B |
GLM-5 MoE(40B active) | Z.ai | 744B |
DeepSeek-V3 0324 MoE(37B active) | DeepSeek | 685B |
Mistral Large 3 MoE(41B active) | Mistral AI | 675B |
DeepSeek-V3.1 Terminus MoE(37B active) | DeepSeek | 671B |
DeepSeek-V3.2 Exp MoE(37B active) | DeepSeek | 671B |
DeepSeek-V3.1 MoE(37B active) | DeepSeek | 671B |
DeepSeek-R1 0528 MoE(37B active) | DeepSeek | 671B |
DeepSeek-V3.2 MoE(37B active) | DeepSeek | 671B |
DeepSeek-V3 MoE(37B active) | DeepSeek | 671B |
DeepSeek-R1 MoE(37B active) | DeepSeek | 671B |
DeepSeek-V4.1-Flash MoE | DeepSeek | 552B |
DeepSeek V4.1 Flash 1017 MoE(16B active) | DeepSeek | 552B* |
Nemotron 3 Ultra MoE(55B active) | NVIDIA | 550B |
PaLM | 540B | |
Megatron-Turing NLG | NVIDIA | 530B |
Qwen 3 Coder 480B A35B MoE(35B active) | Alibaba | 480B |
MiniMax M1 MoE(45.9B active) | MiniMax | 456B |
MiniMax-01 MoE(45.9B active) | MiniMax | 456B |
MiniMax M3 MoE(23B active) | MiniMax | 428B |
ERNIE 4.5 VL 424B A47B MoE(47B active) | Baidu | 424B |
Hermes 4 405B | Nous Research | 405B |
Llama 3.1 405B | Meta | 405B |
Llama 4 Maverick MoE(17B active) | Meta | 400B |
Trinity Large Thinking MoE(13B active) | Arcee AI | 398B |
Nex-N2-Pro MoE(17B active) | Nex AGI | 397B |
Qwen 3.5 397B A17B MoE(17B active) | Alibaba | 397B |
GLM-4.6 MoE(32B active) | Z.ai | 355B |
GLM-4.5 MoE(32B active) | Z.ai | 355B |
Nemotron-4 340B | NVIDIA | 340B |
PaLM 2 | 340B* | |
GLM-5.3-Flash MoE(18B active) | Z.ai | 320B |
Grok 1 MoE(86B active) | xAI | 314B |
MiMo-V2.5 MoE(15B active) | Xiaomi | 310B |
Hy3 MoE(21B active) | Tencent | 295B |
DeepSeek-V4-Flash MoE(13B active) | DeepSeek | 284B |
Inkling Small MoE(12B active) | Thinking Machines Lab | 276B |
DeepSeek-V2 MoE(21B active) | DeepSeek | 236B |
Qwen 3 235B A22B Instruct 2507 MoE(22B active) | Alibaba | 235B |
Qwen 3 235B A22B MoE(22B active) | Alibaba | 235B |
Qwen 3 235B A22B Thinking 2507 MoE(22B active) | Alibaba | 235B |
Qwen 3 VL 235B A22B MoE(22B active) | Alibaba | 235B |
Qwen 3 VL 235B A22B Thinking MoE(22B active) | Alibaba | 235B |
MiniMax M2.1 MoE(10B active) | MiniMax | 230B |
MiniMax M2 MoE(10B active) | MiniMax | 230B |
Seed 1.6 MoE(23B active) | ByteDance | 230B |
GPT-4o | OpenAI | 200B* |
Step 3.7 Flash MoE(11B active) | StepFun | 198B |
Step 3.5 Flash MoE(11B active) | StepFun | 196.8B |
Falcon 180B | TII | 180B |
BLOOM | BigScience | 176B |
GPT-3 | OpenAI | 175B |
Claude 3.5 Sonnet | Anthropic | 175B* |
OPT-175B | Meta | 175B |
Mixtral 8x22B MoE(39B active) | Mistral AI | 141B |
LaMDA | 137B | |
DBRX MoE(36B active) | Databricks | 132B |
Mistral Medium 3.5 | Mistral AI | 128B |
Qwen 3.8 Flash MoE(6B active) | Alibaba | 125B |
Ling 3.0 Flash MoE(5.1B active) | inclusionAI | 124B |
Mistral Large 2 | Mistral AI | 123B |
Qwen 3.5 122B A10B MoE(10B active) | Alibaba | 122B |
Nemotron 3 Super MoE(12B active) | NVIDIA | 120B |
Mistral Small 4 MoE(6.5B active) | Mistral AI | 119B |
Laguna S 2.1 MoE(8B active) | Poolside | 118B |
GPT OSS 120B MoE(5.1B active) | OpenAI | 117B |
Command A | Cohere | 111B |
Llama 4 Scout MoE(17B active) | Meta | 109B |
GLM-4.6V MoE(12B active) | Z.ai | 106B |
GLM-4.5 Air MoE(12B active) | Z.ai | 106B |
GLM-4.5V MoE(12B active) | Z.ai | 106B |
Command R+ | Cohere | 104B |
Solar Pro 3 MoE(12B active) | Upstage | 102B |
Hunyuan A13B Instruct MoE(13B active) | Tencent | 80B |
Qwen 3 Coder Next MoE(3B active) | Alibaba | 80B |
Qwen 3 Next 80B A3B Instruct MoE(3B active) | Alibaba | 80B |
Qwen 3 Next 80B A3B Thinking MoE(3B active) | Alibaba | 80B |
Qwen 2.5 72B | Alibaba | 72.7B |
Virtuoso Large | Arcee AI | 72B |
Qwen 2.5 VL 72B | Alibaba | 72B |
Hermes 4 70B | Nous Research | 70B |
DeepSeek-R1-Distill-Llama-70B | DeepSeek | 70B |
Claude 3 Sonnet | Anthropic | 70B* |
Llama 3.3 70B | Meta | 70B |
Llama 3.1 70B | Meta | 70B |
Llama 3 70B | Meta | 70B |
Llama 2 70B | Meta | 70B |
Mixtral 8x7B MoE(12.9B active) | Mistral AI | 46.7B |
Falcon 40B | TII | 40B |
Nex-N2-mini MoE(3B active) | Nex AGI | 35B |
Qwen 3.6 35B A3B MoE(3B active) | Alibaba | 35B |
Qwen 3.5 35B A3B MoE(3B active) | Alibaba | 35B |
Yi-34B | 01.AI | 34B |
Laguna XS 2.1 MoE(3B active) | Poolside | 33B |
Qwen 3 32B | Alibaba | 32.8B |
Qwen 3 VL 32B | Alibaba | 32.8B |
Qwen 2.5 Coder 32B | Alibaba | 32.5B |
Qwen 2.5 32B | Alibaba | 32B |
Command R | Cohere | 32B |
Solar Pro 2 | Upstage | 31B |
Gemma 4 31B | 30.7B | |
Qwen 3 30B A3B Instruct 2507 MoE(3.3B active) | Alibaba | 30.5B |
Qwen 3 Coder 30B A3B MoE(3.3B active) | Alibaba | 30.5B |
Qwen 3 30B A3B MoE(3.3B active) | Alibaba | 30.5B |
Qwen 3 30B A3B Thinking 2507 MoE(3.3B active) | Alibaba | 30.5B |
Qwen 3 VL 30B A3B MoE(3.3B active) | Alibaba | 30.5B |
Qwen 3 VL 30B A3B Thinking MoE(3.3B active) | Alibaba | 30.5B |
Nemotron 3 Nano MoE(3.5B active) | NVIDIA | 30B |
Nemotron 3.5 Lightning MoE(3B active) | NVIDIA | 30B |
GLM-4.7-Flash MoE(3B active) | Z.ai | 30B |
Muse Glimmer 30B | Meta | 29.6B |
Gemma 3 27B | 27B | |
Qwen 3.8 27B | Alibaba | 27B |
Qwen 3.5 27B | Alibaba | 27B |
Qwen 3.6 27B | Alibaba | 27B |
Gemma 2 27B | 27B | |
Gemma 4 26B A4B MoE(3.8B active) | 25.2B | |
Mistral Small 3.2 | Mistral AI | 24B |
Mistral Small 3.1 | Mistral AI | 24B |
Voxtral Small 24B | Mistral AI | 24B |
Mistral Small 3 | Mistral AI | 24B |
GPT OSS 20B MoE(3.6B active) | OpenAI | 21B |
Claude 3 Haiku | Anthropic | 20B* |
Qwen 3 14B | Alibaba | 14.8B |
Qwen 2.5 14B | Alibaba | 14B |
Phi-4 | Microsoft | 14B |
Ministral 3 14B | Mistral AI | 13.9B |
Nemotron Nano 2 VL 12B | NVIDIA | 12.6B |
Gemma 3 12B | 12B | |
Mistral Nemo | Mistral AI | 12B |
Solar Mini | Upstage | 10.7B |
Qwen 3.5 9B | Alibaba | 9B |
Gemma 2 9B | 9B | |
Nemotron Nano 2 9B | NVIDIA | 9B |
Ministral 3 8B | Mistral AI | 8.8B |
Qwen 3 8B | Alibaba | 8.2B |
Qwen 3 VL 8B | Alibaba | 8.2B |
Qwen 3 VL 8B Thinking | Alibaba | 8.2B |
Granite 4.1 8B | IBM | 8B |
Gemma 3n E4B | 8B | |
GPT-4o mini | OpenAI | 8B* |
Llama 3.1 8B | Meta | 8B |
Llama 3 8B | Meta | 8B |
Ministral 8B | Mistral AI | 8B |
Qwen 2.5 7B | Alibaba | 7.6B |
Mistral 7B | Mistral AI | 7B |
Command R7B | Cohere | 7B |
Phi-4 Multimodal | Microsoft | 5.6B |
Gemma 3 4B | 4B | |
Ministral 3 3B | Mistral AI | 3.8B |
Phi-4 mini | Microsoft | 3.8B |
Phi-3 mini | Microsoft | 3.8B |
Gemini Nano 2 | 3.3B | |
Llama 3.2 3B | Meta | 3.2B |
Granite 4.0 H Micro | IBM | 3B |
Ministral 3B | Mistral AI | 3B |
Gemma 2 2B | 2B | |
Gemini Nano 1 | 1.8B | |
GPT-2 | OpenAI | 1.5B |
Llama 3.2 1B | Meta | 1.2B |
Qwen 2.5 0.5B | Alibaba | 0.5B |
Parameter sizes of known large language models. Published figures and estimates are identified.
3.2 Memory Needs for Local Models
For local models, memory matters most. All weights have to sit in RAM, even for Mixture-of-Experts models where only part of the network is active per token. The chart converts total parameters into gigabytes at 4.89 bits per weight (Q4_K_M quantization). Context cache and runtime add a few more gigabytes in practice:
gradually.aiThe vendor already ships GPT OSS in its own 4-bit format, so its file size deviates from this calculation.
4. Context Windows: Comparing Provider-Reported Limits
The chart shows the documented context window of eight current models. Llama 4 Scout leads by a wide margin at just over 10 million tokens, while most other top models sit at around one million. Grok 4.7 offers half of that:
gradually.aiThe interactive table combines 339 records from the central dataset. Context limits vary by model, provider, and API endpoint:
Model | Developer | Context Window |
|---|---|---|
| Meta | 10.5M | |
| Alibaba | 10M | |
2M | ||
2M | ||
| xAI | 2M | |
| xAI | 2M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| OpenAI | 1.1M | |
| Moonshot AI | 1M | |
| Meta | 1M | |
| Google DeepMind | 1M | |
1M | ||
1M | ||
1M | ||
1M | ||
1M | ||
1M | ||
1M | ||
1M | ||
1M | ||
1M | ||
1M | ||
| DeepSeek | 1M | |
| Moonshot AI | 1M | |
| Meta | 1M | |
| Meta | 1M | |
| Meta | 1M | |
| Poolside | 1M | |
| OpenAI | 1M | |
| OpenAI | 1M | |
| OpenAI | 1M | |
| Google DeepMind | 1M | |
1M | ||
1M | ||
1M | ||
| xAI | 1M | |
| xAI | 1M | |
| xAI | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| Anthropic | 1M | |
| DeepSeek | 1M | |
| DeepSeek | 1M | |
| DeepSeek | 1M | |
| MiniMax | 1M | |
| Alibaba | 1M | |
| Alibaba | 1M | |
| Alibaba | 1M | |
| Meituan | 1M | |
| Thinking Machines Lab | 1M | |
| Thinking Machines Lab | 1M | |
| Meta | 1M | |
| Alibaba | 1M | |
| Alibaba | 1M | |
| Alibaba | 1M | |
| Alibaba | 1M | |
| Z.ai | 1M | |
| Z.ai | 1M | |
| Z.ai | 1M | |
| Xiaomi | 1M | |
| Xiaomi | 1M | |
| Xiaomi | 1M | |
| MiniMax | 1M | |
| NVIDIA | 1M | |
| NVIDIA | 1M | |
| NVIDIA | 1M | |
| NVIDIA | 1M | |
| Amazon | 1M | |
| Amazon | 1M | |
| Amazon | 1M | |
| MiniMax | 1M | |
| Mistral AI | 524.29K | |
| Upstage | 512K | |
| xAI | 500K | |
| xAI | 500K | |
| xAI | 500K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| OpenAI | 400K | |
| Amazon | 300K | |
| Amazon | 300K | |
| Mistral AI | 262.14K | |
| Mistral AI | 262.14K | |
| Moonshot AI | 262.14K | |
| Moonshot AI | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Poolside | 262.14K | |
| Nex AGI | 262.14K | |
| Nex AGI | 262.14K | |
| Tencent | 262.14K | |
| Tencent | 262.14K | |
| inclusionAI | 262.14K | |
| inclusionAI | 262.14K | |
| Arcee AI | 262.14K | |
262.14K | ||
262.14K | ||
| Mistral AI | 262.14K | |
| Mistral AI | 262.14K | |
| Mistral AI | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Alibaba | 262.14K | |
| Moonshot AI | 262.14K | |
| Moonshot AI | 262.14K | |
| Moonshot AI | 262.14K | |
| xAI | 256K | |
| xAI | 256K | |
| xAI | 256K | |
| Mistral AI | 256K | |
| Mistral AI | 256K | |
| Alibaba | 256K | |
| ByteDance | 256K | |
| ByteDance | 256K | |
| Kuaishou | 256K | |
| Kuaishou | 256K | |
| ByteDance | 256K | |
| ByteDance | 256K | |
| ByteDance | 256K | |
| ByteDance | 256K | |
| ByteDance | 256K | |
| ByteDance | 256K | |
| StepFun | 256K | |
| StepFun | 256K | |
| Cohere | 256K | |
| Cohere | 256K | |
| AI21 Labs | 256K | |
| AI21 Labs | 256K | |
| AI21 Labs | 256K | |
| MiniMax | 245.76K | |
| MiniMax | 204.8K | |
| MiniMax | 204.8K | |
| MiniMax | 204.8K | |
| MiniMax | 204.8K | |
| Z.ai | 204.8K | |
| Z.ai | 204.8K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| Anthropic | 200K | |
| OpenAI | 200K | |
| OpenAI | 200K | |
| OpenAI | 200K | |
| OpenAI | 200K | |
| OpenAI | 200K | |
| Z.ai | 200K | |
| Z.ai | 200K | |
| Perplexity | 200K | |
| Z.ai | 200K | |
| 01.AI | 200K | |
| 01.AI | 200K | |
| DeepSeek | 163.84K | |
| DeepSeek | 163.84K | |
| DeepSeek | 163.84K | |
| Meta | 131.07K | |
| Meta | 131.07K | |
| Meta | 131.07K | |
| Meta | 131.07K | |
| Meta | 131.07K | |
| xAI | 131.07K | |
| Mistral AI | 131.07K | |
| Alibaba | 131.07K | |
| Alibaba | 131.07K | |
| Alibaba | 131.07K | |
| Meta | 131.07K | |
| Upstage | 131.07K | |
| Baidu | 131.07K | |
| IBM | 131.07K | |
| IBM | 131.07K | |
| Arcee AI | 131.07K | |
| Nous Research | 131.07K | |
| Nous Research | 131.07K | |
| OpenAI | 131.07K | |
| OpenAI | 131.07K | |
| Mistral AI | 131.07K | |
| Mistral AI | 131.07K | |
| Moonshot AI | 131.07K | |
| Z.ai | 131.07K | |
| Z.ai | 131.07K | |
| Z.ai | 131.07K | |
| Meta | 128K | |
| Meta | 128K | |
| Meta | 128K | |
128K | ||
128K | ||
128K | ||
| xAI | 128K | |
| OpenAI | 128K | |
| OpenAI | 128K | |
| OpenAI | 128K | |
| OpenAI | 128K | |
| OpenAI | 128K | |
| DeepSeek | 128K | |
| DeepSeek | 128K | |
| DeepSeek | 128K | |
| DeepSeek | 128K | |
| DeepSeek | 128K | |
| DeepSeek | 128K | |
| DeepSeek | 128K | |
| DeepSeek | 128K | |
| DeepSeek | 128K | |
| DeepSeek | 128K | |
| Mistral AI | 128K | |
| Mistral AI | 128K | |
| Mistral AI | 128K | |
| Alibaba | 128K | |
| Alibaba | 128K | |
| Alibaba | 128K | |
| Alibaba | 128K | |
| Alibaba | 128K | |
| Alibaba | 128K | |
| Alibaba | 128K | |
| Alibaba | 128K | |
| Alibaba | 128K | |
| NVIDIA | 128K | |
| NVIDIA | 128K | |
| Cohere | 128K | |
| Perplexity | 128K | |
| Perplexity | 128K | |
| Perplexity | 128K | |
| DeepSeek | 128K | |
| DeepSeek | 128K | |
| Alibaba | 128K | |
| Cohere | 128K | |
| Cohere | 128K | |
| Amazon | 128K | |
| Microsoft | 128K | |
| Microsoft | 128K | |
| Microsoft | 128K | |
| Microsoft | 128K | |
| Microsoft | 128K | |
| Microsoft | 128K | |
| 01.AI | 128K | |
| 01.AI | 128K | |
| Nvidia | 128K | |
| Nvidia | 128K | |
| Nvidia | 128K | |
| Reka | 128K | |
| Reka | 128K | |
| Reka | 128K | |
| Zhipu AI | 128K | |
| Zhipu AI | 128K | |
| Baidu | 128K | |
| Mistral AI | 65.54K | |
| Upstage | 65.54K | |
| Z.ai | 65.54K | |
| Microsoft | 64K | |
| Mistral AI | 32.77K | |
| Mistral AI | 32.77K | |
| Mistral AI | 32.77K | |
| Alibaba | 32.77K | |
| Alibaba | 32.77K | |
| Alibaba | 32.77K | |
| Upstage | 32.77K | |
32.77K | ||
| Alibaba | 32.77K | |
| Alibaba | 32.77K | |
| Alibaba | 32.77K | |
| Alibaba | 32.77K | |
| Alibaba | 32.77K | |
| Microsoft | 32.77K | |
| Databricks | 32.77K | |
32K | ||
| Mistral AI | 32K | |
| 01.AI | 32K | |
| Microsoft | 16.38K | |
| 01.AI | 16K | |
8.19K | ||
8.19K | ||
| OpenAI | 8.19K | |
| AI21 Labs | 8.19K | |
| Zhipu AI | 8.19K | |
| Baidu | 8K | |
| Cohere | 4.1K | |
| Nvidia | 4.1K | |
| Stability AI | 4.1K | |
| Stability AI | 4.1K |
Context windows by model and provider endpoint. Each model links to its source.
Meta lists a supported 10.5M-token context window for Llama 4 Scout and says the model was trained with 256,000 tokens. Qwen-Long handles documents up to 10 million tokens through file upload and references. Alibaba's instructions apply only in mainland China's Beijing region.
OpenAI lists a 1,050,000-token context for GPT-6.1 Sol, while Anthropic lists 1M for Claude Opus 5.5 and 1M for Sonnet 5.5.
Mistral documents a 1-million-token limit for Large 4. The central table separately records 524,288 tokens for the OpenRouter endpoint, a different serving configuration. Mistral Small 4 supports 256K tokens, matching its launch announcement and model card.
According to Google's model page, Gemini 4 Argon has a shared context window of 1,048,576 tokens across input, thinking, and output. Synchronous responses allow up to 65,536 output tokens. Long Decode can use more of the remaining context budget, but input and thinking still count toward the shared limit.
For more on the model families, see our overviews of the Claude models and Gemini models.
5. What Does an LLM Cost? Prices per 1 Million Tokens
The chart compares base input prices for selected API models. Values are per 1 million tokens.
GPT-6 Luna and GPT-6.1 Sol have higher price tiers above 272,000 input tokens. Google lists Gemini 3.8 Flash's introductory rates of $0.75 input and $3.75 output through December 31, 2026, followed by scheduled standard rates of $1.50 and $7.50 from January 1, 2027.
Alibaba lists Qwen 3.8 Flash in Frankfurt at a regional yuan rate of CNY 0.8 input and CNY 2.7 output per million tokens. Cache pricing and other conditions can differ. For an estimate for your use, see the API cost calculator.
6. LLM Performance Head to Head
Three leaderboards measure different strengths. Vals Index scores agent tasks, the Artificial Analysis Intelligence Index rates general model performance, and Vectara's leaderboard tracks the hallucination rate when summarizing.
6.1 Vals Index for Agent Tasks
Vals Index v2.1 combines agent tasks in finance, coding, legal work, and tax, weighting each area by its share of U.S. GDP. The chart shows the snapshot from October 27, 2026. Other models sit between ranks 3 and 7 and are not shown.
Claude Sonnet 5.5 leads the index and Gemini 4 Argon ranks third. Google initially rolled it out only to selected cyber defense teams. Google's current model card lists the Gemini app, API, AI Studio, business offerings, Antigravity, and AI Mode as distribution channels; access depends on the specific product.
The gap between Claude Sonnet 5.5 and Claude Opus 5.5 is 0.67 points, which Vals says is within the usual measurement error of about ±0.9 points.
6.2 Artificial Analysis Intelligence Index
The Artificial Analysis Intelligence Index v5.0 combines several benchmarks into one overall score. The chart shows each model's best listed configuration, retrieved on October 29, 2026. The leaderboard lists several reasoning settings of the same model separately, so they are not repeated here.
Gemini 4 Carbon leads with 59 points, and Claude Opus 5.5 follows one point behind. Vals Index and the Artificial Analysis index order the models differently because they weight different tasks.
6.3 Hallucination Rate When Summarizing
Vectara's Hallucination Leaderboard measures how often a model invents content when summarizing a document. Lower is better:
The selection shows current flagship models from the leaderboard. The lowest rates there belong to small models built for summarization, so the list mixes model classes. GPT-6 Sol and Gemini 4 Argon are 5.6 percentage points apart.
7. Open Source vs. Proprietary
Our catalogue classifies 134 records as proprietary and 164 as openly available. Of those, 159 are listed as open weights and 5 as open source.
For March 2026, the Stanford AI Index reports a 3.3% lead (1,503 versus 1,454 rating points) by the top closed model over the top open model on Chatbot Arena.
Six of the ten highest-ranked Arena models were closed in that snapshot. The report compares this with a 0.5% gap in August 2024. Browse open models in our open-source LLM directory.
The chart shows the license classification for the six developer groups with the most records in our catalogue. Proprietary records extend to the left, open ones to the right.
Anthropic has no bar on the right because all Anthropic entries in the catalogue are proprietary.
8. Knowledge Cutoff: How Current Are the Models?
A knowledge cutoff describes the training data date a provider reports for a model. This selection shows provider-reported training cutoffs for six models:
Anthropic lists June 2026 for Opus 5.5 and Sonnet 5.5. OpenAI lists May 18, 2026 for GPT-6 Luna, April 30, 2026 for GPT-6.1 Sol, April 30, 2026 for GPT-6 Astra, and April 20, 2026 for GPT-6 Sol.
9. Selected Releases and the Catalogue Timeline
The timeline shows selected releases with provider announcements. It runs from GPT-4o, DeepSeek-R1, and GPT-5 through Claude Opus 5.5 and Claude Sonnet 5.5 to Gemini 4 Argon and Mistral Large 4. OpenAI records further launches in its API changelog:
Gemini 4 Argon's saved release date is October 13, the date it became generally available according to the Gemini API release notes.
Spreading every model recorded in the gradually.ai catalog across its release month shows how releases have clustered. The darker a cell, the more models came out that month.
10. Model Status: Active, Deprecated, Legacy
This chart shows our status classification for 102 model records from Anthropic, Google, and OpenAI:
Across these 102 records, the dataset classifies 42 active, 40 deprecated, 7 legacy, 5 with limited access, 2 Pro-exclusive, 1 available through API only, 3 preview, 2 open source. Providers can change status or access, so check current model documentation before integrating a model.
11. Market Position and Conclusion
Bottom line:
For a specific use case, compare the relevant task evaluation, actual input and output rates, context limits, and access together. We also publish separate statistics on OpenAI, Anthropic, Google Gemini, Grok, and DeepSeek.
Frequently Asked Questions
The chart shows selected standard API prices per 1 million input tokens. GPT-6 Luna costs $0.1 per million input and $0.5 per million output tokens up to 272,000 input tokens; higher tiers apply above that threshold.
Alibaba’s Frankfurt pricing table for Global deployment lists CNY 0.8 per million input and CNY 2.7 per million output tokens for Qwen3.8-Flash. Its separate Model Studio pricing table lists $0.113 and $0.382 for Global deployment. Alibaba publishes the prices in both currencies; cache and promotional rates are listed separately. For more providers and tiers, see the LLM price index.
