Skip to main content
ToolPotion
AI Advisor
Submit a Tool (Free)
Top 50 LLMs illustration – ToolPotion.ai

Top 50 LLMs

Ranked by ToolRank — ToolPotion's proprietary traction engine

#1 right nowGPT-5.5 AI Model75Strong
measured
93
avg. ToolRank
35.6/100
combined audience
62.8M
ToolRank

ToolRank is ToolPotion’s proprietary ranking engine.

It continuously evaluates hundreds of signals across the open web — real traffic and popularity, audience and community strength, authority and longevity, adoption and engagement — processing millions of data points to place every tool where it has earned to be.

Every AI type is measured by its own tailored model: what proves traction for a SaaS app is not what proves it for an open-source framework or a creator channel. Missing a platform never counts against a tool — ToolRank judges the evidence that exists.

No paid placement. No self-reported numbers.

Among LLMs, DeepSeek leads by a wide margin, with open-source models like DeepSeek-R1 and an API that accepts OpenAI and Anthropic formats. Pick Code Llama for code-only work and GPT-5.6 for coding plus knowledge work. — Editor's verdict

Quick picks

Best overallDeepSeek#2DeepSeek offers open-source frontier models like DeepSeek-V4, DeepSeek-R1 and DeepSeek-Coder, with token-based API pricing.
Best for codeCode Llama#24Code Llama from Meta generates, completes and infills code, with Python-tuned variants and sizes from 7B to 70B parameters.
Best small modelSmolLM3 Language Model#47SmolLM3 packs dual-mode reasoning, six languages and a 128k-token context into an open 3B-parameter model.
1 of 3
  1. # 1Top PickGPT-5.5 AI ModelGPT-5.5 is OpenAI's latest model designed for complex tasks such as coding, research, and data analysis. It offers enhanced efficiency and intelligence, making it ideal for professional and scientific work.75ToolRankStrong
  2. # 2Runner UpDeepSeekDeepSeek is an AI research company dedicated to developing general artificial intelligence. It offers open-source frontier LLMs like DeepSeek-V4, DeepSeek-R1, and DeepSeek-Coder, enabling users to chat with AI or integrate via API for various applications.72ToolRankStrong
  3. # 3Third PlaceHugging FaceHugging Face is a platform dedicated to advancing and democratizing artificial intelligence through open source and open science. It offers a wide range of AI models and tools for developers and researchers.71ToolRankStrong
  1. 4Command R+ ModelCommand R+ is Cohere's large language model optimized for conversational interaction and long-context tasks. It excels in complex RAG workflows and multi-step tool use, providing high performance for production environments.55ToolRankSolid
  2. 5Kimi K3.1 ProKimi K3.1 Pro is an open-weight, natively multimodal agentic model with 2.8 trillion parameters, designed for advanced applications. It features a 1-million-token context window and delivers significant improvements in coding, 3D creation, financial research, and factual accuracy. This model is the first in the K3.1 series, post-trained with large-scale joint reinforcement learning across diverse domains.54ToolRankSolid
  3. 6Mistral Large 4Mistral Large 4 is a powerful, natively multimodal Mixture-of-Experts model with 1 trillion parameters, designed for complex AI workloads. It excels in instruction following, reasoning, coding, and agentic workflows, offering strong performance in cybersecurity, finance, and law. The model supports 160+ languages and features a 1M context length, making it suitable for diverse applications.53ToolRankSolid
  4. 7Codestral AI PlatformCodestral is an open-weight generative AI model designed for code generation tasks. It supports over 80 programming languages, helping developers write and interact with code efficiently. With a 22B parameter model, it offers advanced code completion and error reduction capabilities.48ToolRankSolid
  5. 8DeepSeek-V4.1-Flash-1017DeepSeek-V4.1-Flash-1017 is an updated version of the DeepSeek model, enhancing reasoning, coding, and visual benchmarks. It reduces task execution time and output length, offering faster and cost-effective solutions for users.48ToolRankSolid
  6. 9Mistral Medium 3.5Mistral Medium 3.5 is a frontier-class multimodal model optimized for agentic and coding use cases. It is released as open weights under a Modified MIT license, offering advanced capabilities for diverse applications.47ToolRankSolid
  7. 10DeepSeek-V4-Pro-0813DeepSeek-V4-Pro-0813 is an advanced language model designed to enhance agentic capabilities and performance in production environments. It builds upon the DeepSeek-V4-Pro (Preview) model with improved decoding and reasoning features, making it competitive with leading proprietary models.47ToolRankSolid
  8. 11Nemotron 3 Nano OmniNemotron 3 Nano Omni is an omni-modal reasoning model by NVIDIA that processes images, video, speech, and text. It offers versatile AI capabilities for diverse applications.47ToolRankSolid
  9. 12NVIDIA Nemotron 3.5 Bolt 74BNVIDIA Nemotron 3.5 Bolt 74B is a large language model designed for customization and post-training. It supports text, image, and video inputs, making it versatile for various applications.46ToolRankSolid
  10. 13Qwen3-Coder-NextQwen3-Coder-Next is an open-weight language model tailored for coding agents and local development. It excels in agentic training, offering strong coding capabilities with reduced inference costs.45ToolRankSolid
  11. 14North Mini CodeNorth Mini Code is a 30B-A3B parameter model designed for code generation and agentic software engineering. Developed by Cohere Labs, it supports tool-use capabilities and offers efficient attention implementation for enhanced coding tasks.45ToolRankSolid
  12. 15Qwen3.8-2.4T-A95BQwen3.8-2.4T-A95B is a causal language model designed for complex tasks, offering improved coding, professional work, and research capabilities. It supports extensive context lengths and integrates with popular frameworks for efficient inference.44ToolRankSolid
  13. 16NVIDIA Nemotron 3.5 Super VLNVIDIA Nemotron 3.5 Super VL is a large language model designed for text, image, and video processing. It features a Mamba-2–Transformer hybrid architecture and is best suited for customization, domain adaptation, and full-precision evaluation.43ToolRankSolid
  14. 17Granite Switch 4.1 8BGranite Switch 4.1 8B is a modular LLM that integrates a base model, multiple adapters, and an encoder to provide natural-language responses, decision-making, and speech-to-text capabilities from a single endpoint.43ToolRankSolid
  15. 18Kimi K2.7 CodeKimi K2.7 Code is a coding-focused agentic model that enhances software engineering workflows. It improves token efficiency and task completion, offering substantial advancements over its predecessor, Kimi K2.6.34ToolRankEmerging
  16. 19Solar Open 2Solar Open 2 is a large language model by Upstage, designed for agentic use cases like office productivity and coding. It features a Hybrid-Attention Mixture-of-Experts architecture, offering efficient inference and multilingual support.33ToolRankEmerging
  17. 20Kolibri 1Kolibri 1 is a multilingual large language model developed by Aleph Alpha. It excels in multi-step reasoning, retrieval-augmented generation, and coding tasks, supporting both German and English languages. The model is designed for complex tasks with a large context window and is available under the Apache 2.0 license.33ToolRankEmerging
  18. 21DeepSeek-V4-Flash-0731DeepSeek-V4-Flash-0731 is an advanced language model that enhances agentic capabilities. It outperforms previous versions and is competitive with top proprietary models, offering improved performance across various benchmarks.31ToolRankEmerging
  19. 22Inkling-SmallInkling-Small is a versatile multimodal model that processes text, image, and audio inputs to generate text outputs. It supports developers in creating AI applications like chatbots and coding assistants, with open weights for research and integration.31ToolRankEmerging
  20. 23Mellum2.1 ThinkingMellum2.1 Thinking is a large language model designed for complex tasks in coding, math, and reasoning. It excels in agentic tasks like repository work and tool usage, thanks to extensive reinforcement learning.31ToolRankEmerging
  21. 24Code LlamaCode Llama is Meta's family of large language models specialized for code, built by further training Llama 2 on code-heavy data. It generates code and natural language about code, completes and infills code, and follows coding instructions in a chat format, with dedicated variants for general code, Python and instruction-following.30ToolRankEmerging
  22. 25Hy3Hy3 is a 295B-parameter Mixture-of-Experts model developed by the Tencent Hy Team. It excels in reasoning, agentic, and long-context tasks, offering reliable performance across various productivity scenarios.30ToolRankEmerging
  23. 26LiquidAI LFM2.5-2.6BLiquidAI LFM2.5-2.6B is a hybrid model designed for on-device deployment, offering efficient inference and agentic capabilities. It supports multiple languages and is optimized for various platforms, making it versatile for diverse applications.30ToolRankEmerging
  24. 27Nemotron-Labs-Audex-2BNemotron-Labs-Audex-2B is a unified audio-text large language model designed for audio tasks such as understanding, recognition, and translation. It supports both thinking and instruct modes, offering a compact and efficient solution for diverse audio and text applications.30ToolRankEmerging
  25. 28AliceAI-Foundation-80B-A3B-BaseAliceAI-Foundation-80B-A3B-Base is a foundational language model with a hybrid architecture and MoE layers, featuring 80 billion parameters. It excels in factual knowledge tasks, especially in Russian, and supports a context length of up to 262,144 tokens.30ToolRankEmerging
  26. 29Spark-X2.5-4BSpark-X2.5-4B is a compact, general-purpose language model designed for efficiency and accessibility. It supports over 200 languages and excels in tasks like conversation, writing, and coding, with a context window of up to 1M tokens.29ToolRankEmerging
  27. 30NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16 is a looped large language model designed for general-purpose reasoning. It features a normalized-GPT architecture with sliding-window attention, offering efficient token processing and memory management.29ToolRankEmerging
  28. 31Nanbeige4.2-3BNanbeige4.2-3B is a compact agentic model designed for strong reasoning and alignment capabilities. It excels in general-agent and code-agent tasks, outperforming larger models while maintaining a small parameter size.29ToolRankEmerging
  29. 32Laguna S 2.1Laguna S 2.1 is a Mixture-of-Experts model designed for agentic coding and long-horizon tasks. It features 118 billion parameters and offers advanced reasoning capabilities, making it suitable for complex software engineering applications.29ToolRankEmerging
  30. 33Agents-A1Agents-A1 is a 35B Mixture-of-Experts model designed to scale agentic abilities across multiple domains. It excels in long-horizon search, engineering, scientific research, and instruction following, offering robust performance against frontier-scale systems.29ToolRankEmerging
  31. 34Hy4 PreviewHy4 Preview is a new-generation Mixture-of-Experts model developed by the Tencent Hy Team. It features 770 billion parameters and is designed for productivity across various domains, including software engineering, game development, and scientific research.29ToolRankEmerging
  32. 35Mage-VLMage-VL is a codec-native, proactive-streaming multimodal foundation model designed for efficient image and video understanding. It leverages a unique visual encoder and a causal decoder to enhance real-time streaming perception while reducing computational demands.28ToolRankEmerging
  33. 36GPT-5.6GPT-5.6 is OpenAI's latest AI model, designed to enhance productivity and efficiency across various tasks. It offers improved performance per dollar, enabling users to achieve more with less, making it ideal for demanding workflows in coding, knowledge work, and cybersecurity.28ToolRankEmerging
  34. 37DeepSeek-V3DeepSeek-V3 is a Mixture-of-Experts language model with 671 billion parameters, designed for efficient inference and cost-effective training. It excels in various benchmarks, showcasing strong performance in natural language processing tasks.27ToolRankEmerging
  35. 38Laguna M.1Laguna M.1 is a 225B parameter Mixture-of-Experts model designed for agentic coding and long-horizon tasks. It features high-capacity expert routing and global attention architecture, offering strong performance on various benchmarks.27ToolRankEmerging
  36. 39Ling-3.0-flash-FinLing-3.0-flash-Fin is a finance-enhanced model developed by Ant Group, integrating financial expertise with efficient inference for long-horizon workflows. It offers source-grounded financial search and multi-document reasoning.27ToolRankEmerging
  37. 40Nex-N2.5-miniNex-N2.5-mini is part of the Nex-AGI's next-generation family of agentic models. It is designed for long-horizon tasks in real-world environments, enhancing capabilities in computer use, web browsing, and visually grounded agentic tasks.26ToolRankEmerging
  38. 41Ling-3.1-flashLing-3.1-flash is an advanced hybrid reasoning model designed for complex, real-world tasks. It features 560 billion total parameters and a 1 million-token context window, enhancing performance across various domains including healthcare, finance, and software engineering.26ToolRankEmerging
  39. 42LiquidAI LFM2.5-24B-A2BLiquidAI LFM2.5-24B-A2B is a 24B-parameter reasoning model optimized for fast, private on-device agents. It features efficient inference and supports multiple languages, making it ideal for agentic workflows and tool use.25ToolRankEmerging
  40. 43ALoDLM-8BALoDLM-8B is a diffusion language model developed by Amazon, designed for generating text and code. It uses adaptive computation to refine latent states, enhancing efficiency and accuracy in language tasks. The model is part of the ALoDLM family, which supports noncommercial research in language modeling and code generation.25ToolRankEmerging
  41. 44InternLMIntern Large Models, developed by Shanghai AI Laboratory, offers open-source LLMs and MLLMs. Their suite includes models like InternVL and InternLM-XComposer, alongside a toolchain for development and application, featuring XTuner for fine-tuning and LMDeploy for deployment.21ToolRankEmerging
  42. 45BloombergGPTBloombergGPT is a large language model with 50 billion parameters, specifically tuned for finance-related tasks. It aims to enhance financial data processing and analysis, providing insights and predictions for professionals in the finance industry.21ToolRankEmerging
  43. 46Kimi-K2-Instruct ModelKimi-K2-Instruct is a state-of-the-art mixture-of-experts language model designed for advanced AI tasks. It excels in reasoning, coding, and tool use, offering robust performance with 32 billion activated parameters.20ToolRankEmerging
  44. 47SmolLM3 Language ModelSmolLM3 is a 3B parameter language model designed to push the boundaries of small models. It supports dual mode reasoning, six languages, and long context processing, offering strong performance at the 3B–4B scale.20ToolRankEmerging
  45. 48MiniMax M3 - Coding & Agentic FrontierMiniMax M3 is an advanced AI model excelling in coding and agentic tasks. It features a 1M context window and native multimodal understanding, powered by the MSA architecture. The model offers open-weight capabilities, making it a versatile tool for developers and researchers.19ToolRankNascent
  46. 49BERT Multilingual ModelsBERT offers two multilingual models: Cased and Uncased, supporting over 100 languages. The Cased model is recommended for non-Latin alphabets and general use, while the Uncased model is an older version. These models are designed for natural language understanding tasks and can be fine-tuned for specific applications.19ToolRankNascent
  47. 50WeVisOne-VLWeVisOne-VL is a unified vision-language model developed by WeChat Vision, Tencent Inc. It integrates object-level and pixel-level perception for tasks like object detection and image matting, designed for research purposes.19ToolRankNascent

Quick picks

Best overallDeepSeek#2DeepSeek offers open-source frontier models like DeepSeek-V4, DeepSeek-R1 and DeepSeek-Coder, with token-based API pricing.
Best for codeCode Llama#24Code Llama from Meta generates, completes and infills code, with Python-tuned variants and sizes from 7B to 70B parameters.
Best small modelSmolLM3 Language Model#47SmolLM3 packs dual-mode reasoning, six languages and a 128k-token context into an open 3B-parameter model.
1 of 3

Editor's round-up

This board ranks large language models, from frontier releases to small open models you can run yourself. DeepSeek leads by a wide margin with a ToolRank of 70. It offers open-source models like DeepSeek-V4, DeepSeek-R1 and DeepSeek-Coder, and its API accepts both OpenAI and Anthropic formats. Code Llama is second at 28, Meta's code-focused family built on Llama 2. GPT-5.6 sits third at 26 as OpenAI's model for coding, knowledge work and cybersecurity.

Open-weight options fill the middle. DeepSeek-V3 is a 671B Mixture-of-Experts model with 37B active parameters and a 128K context. Kimi-K2-Instruct Model scales to 1 trillion total parameters and targets coding and tool use. InternLM from Shanghai AI Laboratory adds multimodal and math reasoning models. GLM-5.2 stretches to a 1M-token context under an MIT license.

Some entries fill a niche. BloombergGPT is a 50 billion parameter model tuned for finance. BERT Multilingual Models cover over 100 languages for classification and sentiment work. SmolLM3 Language Model is the compact choice, with 3B parameters and six languages. Mistral-7B-Instruct-v0.3 and Mixtral-8x22B-Instruct-v0.1 both support function calling.

Choose by where the model will run. An API like DeepSeek's gets you live fastest, while an open-weight model gives you control over hosting.

How to choose LLMs

  1. API or self-hosted

    DeepSeek and GPT-5.6 offer APIs, so you can start quickly. Open-weight models like SmolLM3 and GLM-5.2 let you host the model on your own hardware.

  2. Check the context window

    Long documents need long context. GLM-5.2 handles 1M tokens, DeepSeek-V3 and Kimi-K2-Instruct reach 128K, and Falcon3-10B-Instruct stops at 32K.

  3. Match size to hardware

    Bigger is not always practical. SmolLM3 has 3B parameters, while DeepSeek-V3 has 671B in total, so pick a size your servers can actually serve.

  4. Look for function calling

    Agents need a model that can call external functions. Mistral-7B-Instruct-v0.3 and Mixtral-8x22B-Instruct-v0.1 both list function calling support.

Frequently asked questions

More leaderboards

Top 250 AI AppsTop 250 AI Browser PluginsTop 250 AI GitHub ReposTop 250 AI AgentsTop 250 AI Mobile AppsTop 200 AI YouTube ChannelsAll leaderboards