Awesome Specialized (Open-Weight) SLMs

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.

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! 🚧

Awesome Specialized (Open-Weight) SLMs

Schematron (2025)

Task: Deterministic HTML ➜ schema-valid JSON extraction (long-context)
Models: 3B (cost-efficient) · 8B (quality bump). Context up to 128K
Links:
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
Blog / Results: Launch write-up with enterprise annotation focus 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
Enterprise integrations: Supported across search stacks; paired with rerankers/embeddings in commercial offerings (Elastic integration)

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
- Foundation-Sec-8B-Instruct
Announcement: Cisco Foundation AI blog (open-weight, enterprise focus)
Ecosystem signal: Vendor & partner posts on operational use (e.g., SIEM/SOAR stacks) (Splunk integration)

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
- ShieldGemma-2-4b-it
Release: Listed alongside Gemma 3 in Google/Vertex model releases

Llama Guard 4 (Apr. 2025)

Task: Open-weight safety/guardrail classifier (text & multimodal variants) aligned to enterprise policy taxonomies
Links:
- 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
- GGUF quantizations available
Benchmarks / Release: Codestral 25.08 update + open-weight claim; SOTA SWE-Bench-Verified results
Model overview: Mistral docs (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
Enterprise access: Published in Vertex AI Model Garden (managed enterprise deployment)
Official blog / tooling: Qwen3-Coder announcement + 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
Paper / Methods: NV-Embed paper (latent-attention pooling; two-stage instruction tuning)
Enterprise ecosystem: NVIDIA NeMo docs & 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
Enterprise integrations: AWS Marketplace listing; Elastic Open Inference API support
Product page / Benchmarks: Reranker v2 model page 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