# Tagged decision-models — page 3

> A living map of AI that is out in the world: signals from releases and deployments, the map of concepts and makers they connect, and stories that explain what moved.

- [TypeSafe AI launches Jev, a hosted model for typed software decisions](https://shipped.anth.us/decisions/typesafe-jev/index.md): Jev takes state and typed questions and returns probabilities. Early access via API, $0.042 per million input tokens; weights not stated. (2026-09-15)
- [Mistral AI releases Shieldstral 1.0, a 3B multimodal safety classifier under Apache 2.0](https://shipped.anth.us/decisions/mistral-shieldstral-1-0-3b/index.md): Shieldstral judges text or images against a policy you write and returns an unsafe score. Reported best F1 on several benchmarks; Apache-2.0. (2026-07-16)
- [NVIDIA releases Nemotron 3.5 Content Safety, a 4B multilingual moderation classifier](https://shipped.anth.us/decisions/nvidia-nemotron-3-5-content-safety/index.md): Gemma-3-4B-based, 12 languages, text plus one image, with custom-policy mode. Open licence plus Gemma terms; card dated 2 June 2026. (2026-06-02)
- [IBM releases Granite Guardian 4.1 8B, a judge model for safety, RAG and agent checks](https://shipped.anth.us/decisions/ibm-granite-guardian-4-1-8b/index.md): Apache-2.0 judge on Granite 4.1 8B with a hybrid reasoning mode. Covers harm, groundedness and function-calling hallucination, plus custom criteria. (2026-04-16)
- [NVIDIA releases Nemotron 3 Content Safety, a 4B moderation classifier on Gemma-3](https://shipped.anth.us/decisions/nvidia-nemotron-3-content-safety/index.md): Predecessor to 3.5: 4B, 12 languages, 23 safety categories. Released 16 March 2026 under the NVIDIA open model licence plus Gemma terms. (2026-03-16)
- [OpenAI releases gpt-oss-safeguard, open-weight safety reasoning models that read your policy](https://shipped.anth.us/decisions/openai-gpt-oss-safeguard/index.md): 20b and 120b Apache-2.0 models that classify text against a policy you supply at inference time, with configurable reasoning effort. (2025-10-29)
- [Qwen releases Qwen3Guard, multilingual moderation models in 0.6B, 4B and 8B](https://shipped.anth.us/decisions/qwen-qwen3guard/index.md): Apache-2.0 safety classifiers covering 119 languages with Safe, Controversial and Unsafe levels; Stream variants are also on the Hub. (2025-09-23)
- [Skywork releases Reward-V2, eight reward models from 0.6B to 8B](https://shipped.anth.us/decisions/skywork-reward-v2/index.md): For context: Llama and Qwen3-based preference scorers trained on about 26 million pairs. The 8B card reports 96.4 on RewardBench v1. (2025-07-02)
- [Katanemo releases Arch-Router 1.5B, a router that follows your own route descriptions](https://shipped.anth.us/decisions/katanemo-arch-router-1-5b/index.md): For context: a Qwen2.5-1.5B-based model that maps a query to a user-defined route and returns JSON. Released under the Katanemo licence. (2025-06-19)
- [Meta releases Llama Guard 4, a 12B multimodal safety classifier](https://shipped.anth.us/decisions/meta-llama-guard-4-12b/index.md): For context: a dense 12B pruned from Llama 4 Scout that screens prompts and responses across 14 MLCommons hazard categories. Gated under the Llama 4 licence. (2025-04-23)

Page 3 of 3 · [previous](https://shipped.anth.us/tags/decision-models/page/2/index.md)