Text2026
Ricursive Intelligence · USA
A dense (not MoE) model with 65B parameters, pretrained on 12 trillion tokens, with a context of up to 1 million tokens. It is built for programming and agents: by the authors' own test it scores 80.52 on Terminal-Bench 2.1 and beats much larger American open models (Reflection Beam 501B with 80.1, Meta Muse Spark Open 816B with 76.4, NVIDIA Nemotron 3 Ultra 550B with 56.4). The authors started from the Olmo 3 recipe and claim a 16 to 55 times gain in training efficiency on code. It was designed for NVIDIA graphics cards. BF16 and NVFP4 are published for the instruction-tuned version, plus a BF16 base. This is a preview release, and the figures are the authors' own, not independent.
- an agent for programming in the terminal and a repository
- long tasks with a context of hundreds of thousands of tokens
- your own tuning on top of an open base model
- Sizes
- 65B (dense)
- Hardware
- from: Cluster
TextGGUF2025–2026
Moonshot AI · China
Very large Moonshot MoE models for agentic work. K3 (2.8 trillion parameters) was the largest open model at release, with up to 1M tokens of context and image understanding; K2.7-Code is built for programming. On November 4 the weights of K3.1 Pro were opened (2.8 trillion parameters, 104B active, 1M tokens of context, images): a refinement of K3 for programming, 3D and financial research; by the tests of Moonshot itself, Terminal-Bench 4.0 is 65.7%. The weights are in MXFP4, about 1.56 TB, run through vLLM and SGLang, and need a serious cluster.
- Multi-step agents: search, data collection, reports
- In-depth document analysis
- Help for developers
- Sizes
- 16B-A3B – 2.8T-A104B
- Hardware
- from: 1 GPU
Text2026
Nanbeige · China
A 33B model that the authors call the senior teacher for future lightweight agents: they plan to train small models (for example 3B) on it. It is strong in programming and office tasks: on the JobBench test across 35 professions it beats the baseline in most, according to the authors. Context up to 512 thousand tokens, reasoning can be switched on and off, and tool calling works best in the XML format. The authors compare it with Qwen3.8-27B, Muse-Glimmer-30B and Gemma 4 31B and say it leads on most tests; the figures are their own. A base version is published too.
- an agent for programming and working with tools
- office tasks by profession: reports, tables, document analysis
- a base model for your own fine-tuning
- Sizes
- 33B
- Hardware
- from: 1 GPU
Text2026
XHToken (the Spark project) · not stated
A compact 4B model to run right on a device, focused on agents: tool calling, multi-step planning, programming, long context. Version 2.6 came about a month after 2.5 with the same architecture and improved only through further training. The reasoning budget is adjustable: none, low, high. The authors state that it holds its own against models of around 9B. Languages on the card: English and Chinese, Russian is not stated. Running it needs the model's own code in transformers. The figures come from the authors. At the end of October a 1.7B size was added (INT4, INT8, FP8).
- an agent with tool calling on a home computer or laptop
- a local programmer assistant that does not need a big server
- AI built into an app with a choice of how much the model thinks
- Sizes
- 1.7B и 4B
- Hardware
- from: Laptop
Code2026
Recursive · not stated
An open expert model (39.8B parameters, 2.5B active per token) for coding agents: it digs into someone else's repository, fixes bugs, writes functions and tests, and can reason and call tools over several steps. Trained from scratch, 262 thousand token context, text only. It weighs about 80 GB, and the authors run it on a single B300 or H200 card through SGLang. By the authors' own tests: Terminal-Bench 2.1 54.9%, SWE-bench Pro v2 89.7%, SWE-bench Verified 74.2%. Part of the comparisons come from other developers' own reports, and the test conditions differ between sources.
- an agent for fixing bugs and extending a repository in the terminal
- writing tests and new functions from a description
- a local developer assistant on your own server
- Sizes
- 39.8B-A2.5B
- Hardware
- from: 1 GPU
CodeOllama2026
DeepReinforce · not disclosed
Models for agentic development: they build their own plan and scaffolding for a task and execute it in the terminal. Fine-tuned from Qwen 3.5 and Gemma 4; work with Claude Code, OpenHands and similar tools. In October Ornith-1.5-122B (MoE, MIT) came out: according to the authors it is stronger and faster than the previous ones, writes shorter and is meant for a single DGX-class card; compressed FP8, NVFP4, GGUF and MLX versions exist. The figures come from the authors.
- A developer agent in the terminal
- Fixing bugs from a task description
- Understanding and extending a large repository
- Sizes
- 9B – 397B
- Hardware
- from: 1 GPU
Text2026
Reflection AI · USA
An open model from the American company Reflection with 501B parameters (23B active, a 1 million token context), text only, mainly English. Stated as economical in reasoning: according to the authors it needs about three times fewer tokens for thinking at comparable quality. Strong in programming, tool calling and science. Weights in BF16, FP8 and NVFP4. The benchmark figures are the authors' own.
- an agent for solving tasks in a repository and a terminal
- long tasks with tools and search (via MCP)
- solving scientific and mathematical problems with reasoning
- Sizes
- 501B-A23B
- Hardware
- from: Cluster
Code2026
Microsoft · USA
A compact agent for working with repository code, based on Qwen3.5-4B: searches the project, edits files, runs tests and prepares a patch. Trained only with reinforcement learning on about 1500 synthetic tasks. Works through the light Leaf harness with five tools.
- fixing bugs in a repository from a task description
- navigating an unfamiliar codebase
- preparing patches with test runs
- Sizes
- 4B
- Hardware
- from: Laptop
CodeGGUF2025–2026
JetBrains · Czech Republic
JetBrains models for fast code autocompletion. Mellum2 (12B, 2.5B active) is already a full assistant: it writes and edits code, calls tools and reasons. Version 2.1 spent most of its training on reinforcement learning in real repositories with a shell and tests: 47 on SWE-bench Verified, self-reported by JetBrains.
- Fast code autocompletion on your own server
- A developer assistant that does not send code to the cloud
- Fine-tuning on the company's code
- Sizes
- 4B – 12B-A2.5B
- Hardware
- from: Laptop
TextRUOllama2024–2026
Cohere · Canada
Business models: document search with source citations, tool calling, many languages. Command A+ (2026) was the first under Apache 2.0, followed by the North line: code, translation and compact vision.
- Knowledge-base answers with source citations
- Agents that work with internal systems
- Translation and correspondence in different languages
- Sizes
- 2.5B – 218B-A25B
- Hardware
- from: Laptop
Code2025–2026
Kwaipilot (Kuaishou) · China
Kuaishou models for agentic development, trained to solve real tasks in repositories. KAT-Coder-V2.5-Dev (35B, 3B active) is the open version of their closed flagship.
- An agent that fixes tasks in the repository
- Code generation and refactoring
- Automating routine development tasks
- Sizes
- 32B – 72B, 35B-A3B
- Hardware
- from: 1 GPU
CodeGGUF2026
Poolside · USA
Models for agentic programming: they edit code in a repository on their own. The small XS runs on a Mac with 36 GB of memory; S 2.1 has a 1M-token context.
- Coding agent for in-house development
- Bug fixing and code improvements
- Working with large codebases
- Sizes
- 33B-A3B – 225B-A23B
- Hardware
- from: 1 GPU
TextRUOllama2023–2026
Mistral AI · France
European models focused on speed. Mixtral was one of the first open mixture-of-experts models; there are versions for images (Pixtral, Medium 3.5), Lean proofs and moderation (Shieldstral).
- Fast chat responses
- Data extraction from text
- Translation and multilingual work
- Sizes
- 3B – 675B
- Hardware
- from: Laptop
TextOllama2024–2026
Google · USA
Compact Google models that run well on a single computer; larger versions understand images. Includes CodeGemma for code, FunctionGemma 270M for function calling and the fast DiffusionGemma.
- Offline assistant on a laptop
- Reading photos of documents and receipts
- Customer request classification
- Sizes
- 270M – 31B
- Hardware
- from: Laptop
Drawings and CADGGUF2026
Autodesk Research · USA
A model from Autodesk Research: from eight rendered views of a part it writes an executable CadQuery program. It was trained on synthetic data only. It produces a draft; an engineer checks and signs off the document.
- Getting a parametric model from a set of views of a part
- Converting finished geometry into editable code
- A draft model for a design office
- Sizes
- 2B (based on Qwen3-VL-2B)
- Hardware
- from: Laptop
CodeOllama2025–2026
Essential AI · USA
An 8B model trained from scratch by the company of one of the authors of the transformer architecture. Strong at code and technical tasks; version 1.5 handles context up to 160K tokens.
- Writing and fixing code
- A developer agent on a single GPU
- Solving technical and scientific problems
- Sizes
- 8B
- Hardware
- from: Laptop
Math and reasoningGGUF2025–2026
Princeton University · USA
Open models for formal proofs in Lean 4 from Princeton. The new Goedel-Code-Prover proves program correctness.
- Formal verification of mathematical workings
- Verifying code correctness
- Training
- Sizes
- 7B – 32B
- Hardware
- from: Laptop
CodeGGUF2026
IQuest Research · China
A family of coding models with standard and reasoning versions, including a Loop variant that runs through its layers a second time. Sizes from 7B to 40B.
- Writing and refining code
- Solving tasks with step-by-step reasoning
- Agentic work with a repository
- Sizes
- 7B – 40B
- Hardware
- from: Laptop
CodeOllama2024–2026
Alibaba (Qwen team) · China
The broadest open coding family: from 0.5B for autocompletion to 480B for agents. Qwen3-Coder-Next (80B, 3B active) works as a developer agent on a single GPU.
- Code autocompletion in the editor
- An agent that edits code in the repository on its own
- Writing and refining scripts, SQL and integrations
- Sizes
- 0.5B – 480B-A35B
- Hardware
- from: Laptop
Code2026
Ai2 (Allen Institute for AI) · USA
Fully open developer agents from Ai2: weights, data and training recipe are all public. Designed so a company can cheaply fine-tune the agent on its own repository.
- An agent for fixing issues in code
- Fine-tuning the agent on an internal repository
- Automating small edits and tests
- Sizes
- 8B – 32B
- Hardware
- from: Laptop
CodeRUOllama2025
Mistral AI (with All Hands AI) · France
Mistral models for agentic development: they read the repository, edit files and run commands on their own. The 24B version fits on a single GPU.
- A developer agent that fixes tickets from the tracker
- Extending internal systems from a description
- Automating routine code edits
- Sizes
- 24B – 123B
- Hardware
- from: 1 GPU
Text to SQLGGUF2024–2025
Prem AI · UK
A text-to-SQL model of just 1B parameters, designed to run locally so the database never leaves for external services.
- Local translation of questions into SQL with no internet access
- Query hints on modest hardware
- Embedding into internal analytics tools
- Sizes
- 1B
- Hardware
- from: Laptop
Math and reasoningGGUF2025
Moonshot AI and Project Numina · China, France
Models for formal proofs in Lean 4 from Moonshot AI (Kimi) and Numina. Small versions from 0.6B run on a laptop.
- Formal verification of mathematical workings
- Translating a problem from plain language into Lean
- Training and olympiad preparation
- Sizes
- 0.6B – 72B
- Hardware
- from: Laptop
Math and reasoningOllama2025
Agentica (Berkeley, Sky Computing Lab) and Together AI · USA
Small models fine-tuned with reinforcement learning: DeepScaleR (1.5B) solves olympiad maths, DeepCoder writes code, DeepSWE works as a developer agent. Recipes and data are open.
- Solving maths problems with step-by-step working
- Generating and checking code
- An agent for fixing bugs in a repository
- Sizes
- 1.5B – 32B
- Hardware
- from: Laptop
Text to SQL2025
Snowflake · USA
A Snowflake model for turning questions into SQL, trained with reinforcement learning by checking query results. The open 7B version is based on Qwen2.5-Coder.
- Plain-language questions to a data warehouse
- Generating SQL for reports and dashboards
- Checking and fixing analysts' queries
- Sizes
- 7B
- Hardware
- from: Laptop
CodeRU2025
MTS AI (MWS AI) · Russia
A small coding assistant from MTS AI that understands requests in Russian. Runs locally, with plugins for VS Code and JetBrains.
- Code suggestions and completion in the editor
- Code explanations in Russian
- Drafts of tests and documentation
- Sizes
- 1.5B
- Hardware
- from: Laptop
CodeGGUF2025
ByteDance Seed · China
A compact 8B coding model from ByteDance in base, instruct and reasoning versions. Its training data was selected by the model itself, with almost no hand-written rules.
- Code autocompletion and generation
- Solving algorithmic problems
- A base for fine-tuning on your own stack
- Sizes
- 8B
- Hardware
- from: Laptop
Math and reasoningGGUF2024–2025
DeepSeek · China
DeepSeek models for formal proofs in Lean 4: the proof is checked by a program, not a person. A narrow tool for mathematicians and engineers.
- Formal verification of mathematical workings
- Verifying algorithm correctness
- Training and olympiad preparation
- Sizes
- 7B – 671B
- Hardware
- from: Laptop
CybersecurityGGUF2023–2025
Kindo · USA
One of the best-known open families for security and DevSecOps work: reviewing code for weaknesses, test scenarios, explaining attacks.
- Finding weak spots in code and configurations
- Reviewing incidents and explaining attack techniques
- Drafting scripts and procedures for the security team
- Sizes
- 7B – 70B
- Hardware
- from: Laptop
Text to SQL2025
Alibaba · China
Alibaba models for turning questions into SQL, based on Qwen2.5-Coder. They work with different SQL dialects; a small 3B version suits modest hardware.
- Plain-language database questions
- Queries for different databases (PostgreSQL, MySQL, SQLite)
- Automating routine reports
- Sizes
- 3B – 32B
- Hardware
- from: Laptop
Drawings and CAD2024–2025
University of Luxembourg (SnT) and Artec3D · Luxembourg
Turns a point cloud of a part into executable CadQuery Python code, that is, a parametric 3D model you can open and edit in CAD. The output is a draft: an engineer checks and signs off the document.
- Recovering a parametric model from a 3D scan of a part
- Converting a point cloud into editable CadQuery code
- Preparing a draft model for a design engineer to finish
- Sizes
- 1.5B (Qwen2-1.5B plus a single projection layer)
- Hardware
- from: Laptop
Text to SQL2025
Renmin University of China (RUC) · China
Models for turning questions into SQL, trained on millions of synthetic query examples across different databases. Three sizes for different hardware.
- Database questions without knowing SQL
- Generating queries for reports
- A base for fine-tuning on your own database schema
- Sizes
- 7B – 32B
- Hardware
- from: Laptop
Drawings and CADGGUF2025
ServiceNow Research, Mila and ETS Montreal · Canada
It turns a raster image into SVG code: instead of tracing pixels the model writes vector commands, so lines and symbols stay editable. It simplifies complex drawings, so the result needs a visual check.
- Vectorising diagrams, symbols and logos from raster images
- Turning scans of simple drawings into editable vectors
- Recovering icons and pictograms from images
- Sizes
- 1B and 8B
- Hardware
- from: Laptop
CodeOllama2024
INF Technology · China
Fully reproducible coding models: along with the weights, the data, its cleaning pipeline and the training recipe are open. Understand English and Chinese.
- Code generation and completion
- Training your own coding model from an open recipe
- A programming assistant on low-end hardware
- Sizes
- 1.5B – 8B
- Hardware
- from: Laptop
CodeOllama2023–2024
DeepSeek · China
DeepSeek's coding model family: from small autocompletion models to the large MoE V2, which matched closed models in 2024. Later, coding moved into DeepSeek's general models.
- Code autocompletion and generation
- Translating code between programming languages
- Finding bugs and explaining other people's code
- Sizes
- 1.3B – 236B-A21B
- Hardware
- from: Laptop
Tabular dataGGUF2024
RUCKBReasoning, Renmin University of China · China
A model for office work with tables: for a given question it returns either a direct answer or code to process the data in a table or document.
- Processing tables from Excel and documents from a text instruction
- Generating code for recalculations and selections
- Answering questions about data in reports
- Sizes
- 7B и 13B
- Hardware
- from: Laptop
CodeOllama2024
01.AI · China
Coding models from 01.AI at 1.5B and 9B with a 128K-token context and support for 52 programming languages. A separate line next to the text Yi models.
- Code autocompletion and generation
- Explaining and refactoring code
- A programming assistant without the cloud
- Sizes
- 1.5B – 9B
- Hardware
- from: Laptop
TextOllamaNot maintained2023–2024
Stability AI · UK
Small models from Stability AI: StableLM 2 (1.6B) knows 7 European languages, Stable Code (3B) completes code. No updates since 2024.
- A lightweight chatbot on an ordinary PC
- Code autocompletion in the editor
- A base for fine-tuning on your own task
- Sizes
- 1.6B – 12B
- Hardware
- from: Laptop
CodeOllamaNot maintained2023–2024
Zhipu AI (Z.ai) and Tsinghua University · China
Coding models from the creators of GLM. CodeGeeX4-ALL-9B, based on GLM-4-9B, combines autocompletion, code chat, function calling and repository search in one model.
- Code autocompletion in the IDE
- A code chat assistant
- Answering questions about a repository
- Sizes
- 6B – 9B
- Hardware
- from: Laptop
CodeOllamaNot maintained2024
Mistral AI · France
Mistral's coding model covering 80+ programming languages. The open weights of the main version cannot be used in production without a paid license; newer Codestral versions are API-only.
- Evaluation and testing before buying a license
- Code autocompletion (with a commercial license)
- Research on coding model quality
- Sizes
- 7B – 22B
- Hardware
- from: Laptop
TextOllamaNot maintained2023–2024
OpenChat (Tsinghua University) · China
Fine-tunes of Mistral 7B and Llama 3 8B using the C-RLFT method that caught up with ChatGPT-3.5 in 2023–2024 at just 7–8B. A lightweight general-purpose assistant for a modest server.
- Chat assistant on an inexpensive server
- Drafts of emails and replies
- Help with simple code
- Sizes
- 7B – 13B
- Hardware
- from: Laptop
Text to SQLOllamaNot maintained2023–2024
Defog · USA
One of the first open models that turn a plain-language question into an SQL query against a database. Available in Ollama, but newer competitors are already stronger.
- Answering managers' questions from the sales database without an analyst
- Drafting SQL queries for reports
- An assistant inside a BI system
- Sizes
- 7B – 70B
- Hardware
- from: Laptop
Text to SQLGGUFNot maintained2024
Chat2DB · China
A text-to-SQL model from the open Chat2DB database client: it supports different SQL dialects, with an English and Chinese model card.
- Turning a question into SQL inside a database client
- Drafting queries for different database engines
- Hints for developers working with a schema
- Sizes
- 7B
- Hardware
- from: Laptop
CodeOllamaNot maintained2023–2024
BigCode (Hugging Face and ServiceNow) · USA / France
One of the first open coding models, trained on an open set of source code with an option to exclude your own repository. Today it is more a base for fine-tuning than a leader.
- Code autocompletion in the editor
- Fine-tuning on the company's internal code
- Generating boilerplate code and tests
- Sizes
- 1B – 15B
- Hardware
- from: Laptop
Text to SQLGGUFNot maintained2024
ChatDB · USA
A text-to-SQL model built on DeepSeek-Coder, aimed at complex questions spanning several tables and conditions.
- Complex queries joining several tables
- Answering database questions without an analyst
- Drafting SQL for reports and exports
- Sizes
- 7B
- Hardware
- from: Laptop
CodeOllamaNot maintained2023–2024
Meta · USA
A version of Llama 2 further trained on code, with variants for Python and for chat. Outdated, but many ready-made fine-tuned versions and tools exist.
- Code autocompletion and explanation
- Generating Python scripts
- Base model for fine-tuning on your own stack
- Sizes
- 7B – 70B
- Hardware
- from: Laptop
Text to SQLOllamaNot maintained2024
MotherDuck and Numbers Station · USA
A model for turning questions into SQL, built for the embedded analytics database DuckDB. Available in Ollama, convenient for working with CSV and Parquet locally.
- Plain-language questions about CSV and Parquet exports
- DuckDB queries inside analytics scripts
- Quick analytics on a laptop without a server
- Sizes
- 7B
- Hardware
- from: Laptop
TextOllamaNot maintained2023–2024
WizardLM (Microsoft and Peking University) · USA / China
Fine-tunes of Llama, Mistral and StarCoder using Evol-Instruct, which automatically makes instructions more complex. WizardLM-2 was released in April 2024 and removed almost immediately, so only the 2023 versions are relevant.
- Complex multi-step instructions
- Help for developers
- Solving math problems
- Sizes
- 7B – 70B
- Hardware
- from: Laptop
Text to SQLNot maintained2023
RUCKBReasoning, Renmin University of China · China
An early line of open text-to-SQL models starting at 1B, including variants fine-tuned for specific database schemas.
- Turning an employee question into an SQL query
- Drafting warehouse queries for a report
- Embedding into a BI dashboard as a helper
- Sizes
- 1B – 15B
- Hardware
- from: Laptop
Text to SQLGGUFNot maintained2023
Numbers Station · USA
One of the first open text-to-SQL lines, including very small versions from 350M that run on an ordinary PC.
- Turning a question into SQL from a table description
- Hints while writing queries
- A local analyst helper with no data leaving the company
- Sizes
- 350M – 7B
- Hardware
- from: Laptop
