mirror of
https://github.com/context-labs/awesome-open-workhorse-models.git
synced 2026-10-17 13:03:44 -05:00
adds table
This commit is contained in:
mike
@@ -1,125 +1,31 @@
|
||||
# Awesome Specialized (Open-Weight) SLMs [](https://awesome.re)
|
||||
# Open Workhorse Models
|
||||
|
||||
Purpose-built, **open-weight** small language (and multimodal) models that enterprises actually use for **one** thing—and do it extremely well. No general-purpose LLMs, just models that are extremely good at a specific task and are production ready.
|
||||
[](https://awesome.re)
|
||||
|
||||
Purpose-built, **open-weight** small language (and multimodal) models that enterprises actually use for **one** thing—and do it extremely well. No general-purpose LLMs, no pre-2025 releases.
|
||||
|
||||
> As enterprises productionize AI, narrowly focused SLMs often beat general models on **accuracy, latency, cost, and control** for a given task (extraction, safety, coding agents, search). This list highlights those specialists with real-world traction and open weights.
|
||||
|
||||
🚧 Missed something enterprise-grade? Open a PR! 🚧
|
||||
|
||||
---
|
||||
|
||||
## Table of Contents
|
||||
|
||||
- [Awesome Specialized (Open-Weight) SLMs](#awesome-specialized-open-weight-slms-)
|
||||
- [Schematron (2025)](#schematron-2025)
|
||||
- [ClipTagger-12B (Aug. 2025)](#cliptagger-12b-aug-2025)
|
||||
- [ReaderLM-v2 (Jul. 2025)](#readerlm-v2-jul-2025)
|
||||
- [Foundation-Sec-8B (Cisco) (Apr.–Aug. 2025)](#foundation-sec-8b-cisco-apr-aug-2025)
|
||||
- [ShieldGemma 2 (Mar. 2025)](#shieldgemma-2-mar-2025)
|
||||
- [Llama Guard 4 (Apr. 2025)](#llama-guard-4-apr-2025)
|
||||
- [Devstral Small (24B) — Codestral 25.08 family (May–Jul. 2025)](#devstral-small-24b--codestral-2508-family-may-jul-2025)
|
||||
- [Qwen3-Coder (Jul.–Sep. 2025)](#qwen3-coder-jul-sep-2025)
|
||||
- [NV-Embed-v2 (Sep. 2025)](#nv-embed-v2-sep-2025)
|
||||
- [Jina Reranker v2 (Base-Multilingual) (Jul. 2025 refresh)](#jina-reranker-v2-base-multilingual-jul-2025-refresh)
|
||||
- [What qualifies for this list?](#what-qualifies-for-this-list)
|
||||
- [Contributing](#contributing)
|
||||
|
||||
---
|
||||
|
||||
## Schematron (2025)
|
||||
|
||||
- **Task**: Deterministic HTML ➜ schema-valid JSON extraction (long-context)
|
||||
- **Models**: 3B (cost-efficient) · 8B (quality bump). Context up to 128K
|
||||
- **Links**:
|
||||
- [Schematron-3B](https://huggingface.co/inference-net/Schematron-3B)
|
||||
- [Schematron-8B](https://huggingface.co/inference-net/Schematron-8B)
|
||||
- [Ollama card](https://ollama.com/Inference/Schematron)
|
||||
- **Enterprise use**: Schema-first extraction with strict JSON outputs; built for web data pipelines and RAG pre-structuring
|
||||
|
||||
## ClipTagger-12B (Aug. 2025)
|
||||
|
||||
- **Task**: High-throughput video/frame tagging with structured JSON per keyframe/clip
|
||||
- **Links**:
|
||||
- [ClipTagger-12B](https://huggingface.co/inference-net/ClipTagger-12b)
|
||||
- **Blog / Results**: [Launch write-up with enterprise annotation focus](https://inference.net/blog/cliptagger-12b) and cost/perf vs. closed models
|
||||
|
||||
## ReaderLM-v2 (Jul. 2025)
|
||||
|
||||
- **Task**: Long-context HTML reading ➜ Markdown/JSON; layout-aware conversion for crawling & document pipelines
|
||||
- **Links**:
|
||||
- [ReaderLM-v2](https://huggingface.co/jinaai/ReaderLM-v2)
|
||||
- **Enterprise integrations**: Supported across search stacks; paired with rerankers/embeddings in commercial offerings ([Elastic integration](https://www.elastic.co/search-labs/blog/jina-ai-embeddings-rerank-model-open-inference-api))
|
||||
|
||||
## Foundation-Sec-8B (Cisco) (Apr.–Aug. 2025)
|
||||
|
||||
- **Task**: Cybersecurity LLM for SOC workflows: log triage, CTI normalization, incident summaries, secure-by-default code assist
|
||||
- **Links**:
|
||||
- [Foundation-Sec-8B](https://huggingface.co/fdtn-ai/Foundation-Sec-8B)
|
||||
- [Foundation-Sec-8B-Instruct](https://huggingface.co/fdtn-ai/Foundation-Sec-8B-Instruct)
|
||||
- **Announcement**: [Cisco Foundation AI blog](https://blogs.cisco.com/security/foundation-sec-cisco-foundation-ai-first-open-source-security-model) (open-weight, enterprise focus)
|
||||
- **Ecosystem signal**: Vendor & partner posts on operational use (e.g., SIEM/SOAR stacks) ([Splunk integration](https://www.splunk.com/en_us/blog/artificial-intelligence/accelerating-security-operations-with-splunk-and-foundation-ai-s-first-open-source-security-model.html))
|
||||
|
||||
## ShieldGemma 2 (Mar. 2025)
|
||||
|
||||
- **Task**: Multimodal safety classifier (image+text) for pre/post-moderation—NSFW, violence, dangerous content, hate
|
||||
- **Links**:
|
||||
- [ShieldGemma 2 in Transformers docs](https://huggingface.co/docs/transformers/model_doc/shieldgemma2)
|
||||
- [ShieldGemma-2-4b-it](https://huggingface.co/google/shieldgemma-2-4b-it)
|
||||
- **Release**: Listed alongside Gemma 3 in [Google/Vertex model releases](https://cloud.google.com/vertex-ai/generative-ai/docs/release-notes)
|
||||
|
||||
## Llama Guard 4 (Apr. 2025)
|
||||
|
||||
- **Task**: Open-weight safety/guardrail classifier (text & multimodal variants) aligned to enterprise policy taxonomies
|
||||
- **Links**:
|
||||
- [Llama-Guard-4-12B](https://huggingface.co/meta-llama/Llama-Guard-4-12B)
|
||||
- **Enterprise access**: HF providers listed for turnkey inference
|
||||
|
||||
## Devstral Small (24B) — Codestral 25.08 family (May–Jul. 2025)
|
||||
|
||||
- **Task**: Agentic coding—multi-file edits, tool use, SWE-Bench Verified SOTA among open models
|
||||
- **Links**:
|
||||
- [Devstral-Small-2505/2507](https://huggingface.co/mistralai/Devstral-Small-2507)
|
||||
- GGUF quantizations available
|
||||
- **Benchmarks / Release**: [Codestral 25.08 update](https://mistral.ai/news/codestral-25-08) + open-weight claim; SOTA SWE-Bench-Verified results
|
||||
- **Model overview**: [Mistral docs](https://docs.mistral.ai/getting-started/models/models_overview/) (Apache-2.0, long context, open-weights)
|
||||
|
||||
## Qwen3-Coder (Jul.–Sep. 2025)
|
||||
|
||||
- **Task**: Agentic code generation & long-horizon editing (extreme context windows; multiple sizes, open weights)
|
||||
- **Links**:
|
||||
- [Qwen3-Coder collection](https://huggingface.co/collections/Qwen/qwen3-coder-687fc861e53c939e52d52d10)
|
||||
- **Enterprise access**: Published in [Vertex AI Model Garden](https://cloud.google.com/vertex-ai/generative-ai/docs/maas/qwen/qwen3-coder) (managed enterprise deployment)
|
||||
- **Official blog / tooling**: [Qwen3-Coder announcement](https://qwenlm.github.io/blog/qwen3-coder/) + CLI/tooling for agentic coding
|
||||
|
||||
## NV-Embed-v2 (Sep. 2025)
|
||||
|
||||
- **Task**: Enterprise text embeddings for RAG—top-tier MTEB score; long-doc retrieval; open weights
|
||||
- **Links**:
|
||||
- [NV-Embed-v2](https://huggingface.co/nvidia/NV-Embed-v2)
|
||||
- **Paper / Methods**: [NV-Embed paper](https://arxiv.org/html/2405.17428v3) (latent-attention pooling; two-stage instruction tuning)
|
||||
- **Enterprise ecosystem**: [NVIDIA NeMo docs](https://docs.nvidia.com/nemo/microservices/25.9.0/fine-tune/models/embedding.html) & vendor rundowns positioning v2 for production RAG
|
||||
|
||||
## Jina Reranker v2 (Base-Multilingual) (Jul. 2025 refresh)
|
||||
|
||||
- **Task**: Cross-encoder reranking for multilingual search/RAG; pairs with NV-Embed/BGE/GTE, etc.
|
||||
- **Links**:
|
||||
- [jina-reranker-v2-base-multilingual](https://huggingface.co/jinaai/jina-reranker-v2-base-multilingual)
|
||||
- **Enterprise integrations**: [AWS Marketplace listing](https://aws.amazon.com/marketplace/pp/prodview-uencv3yyikiyu); [Elastic Open Inference API support](https://www.elastic.co/search-labs/blog/jina-ai-embeddings-rerank-model-open-inference-api)
|
||||
- **Product page / Benchmarks**: [Reranker v2 model page](https://jina.ai/reranker/) for perf & usage notes
|
||||
|
||||
---
|
||||
|
||||
## What qualifies for this list?
|
||||
|
||||
- **Specialized**: Excels at a **narrow, well-defined** task (extraction, safety, coding, embeddings/rerankers, etc.)
|
||||
- **Open weights**: Training code/datasets optional but welcome
|
||||
- **Enterprise adoption**: Public deployments, marketplace entries, enterprise docs, or similar signals
|
||||
- **2025+ only**: Initial release (or major refresh) in **2025 or later**
|
||||
|
||||
---
|
||||
|
||||
## Contributing
|
||||
- **2025+** only: Initial release (or major refresh) in **2025 or later**
|
||||
|
||||
🚧 Missed something enterprise-grade? Open a PR! 🚧
|
||||
|
||||
> PRs should include: links (HF/GitHub/docs), size(s), task, license, enterprise signals, and concrete eval notes or deployment guides.
|
||||
## Open Workhorse Models Index
|
||||
|
||||
| Model | Task | Companies Using It | Weights/Cards | Key Features |
|
||||
|-------|------|-------------------|---------------|--------------|
|
||||
| **Schematron (2025)** | Deterministic HTML ➜ **schema-valid JSON** extraction (long-context) | Financial data aggregators, e-commerce platforms, research firms, content monitoring services | [Schematron-3B (HF)](https://huggingface.co/inference-net/Schematron-3B) · [Ollama card](https://ollama.com/Inference/Schematron) | 3B/8B variants · 128K context · Schema-first extraction · Web data pipelines |
|
||||
| **ClipTagger-12B (Aug. 2025)** | High-throughput **video/frame tagging** with **structured JSON** | Video streaming platforms, content moderation services, digital asset management companies, advertising networks | [ClipTagger-12B (HF)](https://huggingface.co/inference-net/ClipTagger-12b) | 12B multimodal · FP8 optimized · Enterprise annotation · Cost vs. closed models |
|
||||
| **ReaderLM-v2 (Jul. 2025)** | **Long-context HTML reading** ➜ Markdown/JSON; layout-aware conversion | Search engines, content aggregators, web scraping services, document processing platforms | [ReaderLM-v2 (HF)](https://huggingface.co/jinaai/ReaderLM-v2) | Crawling pipelines · Document processing · Search stack integration |
|
||||
| **Foundation-Sec-8B (Cisco) (Apr.–Aug. 2025)** | **Cybersecurity** LLM for SOC workflows: log triage, CTI normalization | Cisco, Splunk, security service providers, enterprise SOC teams, threat intelligence firms | [Foundation-Sec-8B (HF)](https://huggingface.co/fdtn-ai/Foundation-Sec-8B) · [Instruct](https://huggingface.co/fdtn-ai/Foundation-Sec-8B-Instruct) | SOC automation · Incident response · Secure code assist · SIEM/SOAR integration |
|
||||
| **ShieldGemma 2 (Mar. 2025)** | **Multimodal safety classifier** (image+text) for pre/post-moderation | Social media platforms, content moderation services, online marketplaces, gaming companies | [ShieldGemma 2 (HF)](https://huggingface.co/google/shieldgemma-2-4b-it) · [Transformers docs](https://huggingface.co/docs/transformers/model_doc/shieldgemma2) | NSFW detection · Violence classification · Content moderation · Google/Vertex integration |
|
||||
| **Llama Guard 4 (Apr. 2025)** | Open-weight **safety/guardrail** classifier (text & multimodal variants) | Enterprise AI platforms, cloud providers, content platforms, educational institutions | [Llama-Guard-4-12B (HF)](https://huggingface.co/meta-llama/Llama-Guard-4-12B) | Safety guardrails · Policy alignment · Enterprise taxonomy · HF provider support |
|
||||
| **Devstral Small (24B) (May–Jul. 2025)** | **Agentic coding**—multi-file edits, tool use, SWE-Bench Verified SOTA | Enterprise software companies, DevOps platforms, code automation services, AI coding assistants | [Devstral-Small-2505/2507 (HF)](https://huggingface.co/mistralai/Devstral-Small-2507) · GGUF quantizations | SWE-Bench SOTA · Multi-file editing · Tool use · Apache-2.0 license |
|
||||
| **Qwen3-Coder (Jul.–Sep. 2025)** | **Agentic code generation & long-horizon editing** (extreme context windows) | Enterprise software development teams, code generation platforms, AI coding tools, development agencies | [Qwen3-Coder (HF)](https://huggingface.co/collections/Qwen/qwen3-coder-687fc861e53c939e52d52d10) | 480B MoE variant · Long context · CLI tooling · Vertex AI integration |
|
||||
| **NV-Embed-v2 (Sep. 2025)** | **Enterprise text embeddings** for RAG—top-tier **MTEB** score | Enterprise search platforms, RAG systems, document management companies, knowledge bases | [NV-Embed-v2 (HF)](https://huggingface.co/nvidia/NV-Embed-v2) | MTEB leader · Long-doc retrieval · Latent attention · NVIDIA NeMo integration |
|
||||
| **Jina Reranker v2 (Jul. 2025)** | **Cross-encoder reranking** for multilingual search/RAG | Search engines, e-commerce platforms, multilingual content platforms, enterprise search solutions | [jina-reranker-v2-base-multilingual (HF)](https://huggingface.co/jinaai/jina-reranker-v2-base-multilingual) | Multilingual support · AWS Marketplace · Elastic integration · Cross-encoder architecture |
|
||||
Reference in New Issue
Block a user