Segan Subramanian retweeted
🚀 Meet Apriel Voice 27B Preview: an open-weight voice AI built to get work done.
Beyond conversation: adapt to dynamic instructions, reason across modalities, and orchestrate multi-step tool calls to complete real tasks.
See Apriel Voice in action. 👇
Segan Subramanian retweeted
Excited to share that our paper AgentJudgeBench has been accepted to #EMNLP2026 (Main Conference)! 🎉
We study a question that becomes increasingly important as agents get deployed:
Can we trust LLMs to judge other agents?
Paper: arxiv.org/pdf/2608.26623v1
🧵
Segan Subramanian retweeted
What if we didn't need MCP servers after all?
What if we didn't need browser-use agents either?
What if... Claude Code was enough?
In our latest paper, we test exactly this: Can simple terminal agents outperform web agents and tool-based agents on real enterprise tasks?
Segan Subramanian retweeted
🧵 Introducing 𝐄𝐧𝐭𝐞𝐫𝐩𝐫𝐢𝐬𝐞𝐎𝐩𝐬-𝐆𝐲𝐦🚀 : a rigorous new benchmark for stateful agentic planning and tool use in real enterprise environments.
1,150 expert-curated tasks · 512 tools · 164 DB tables · 8 domains. Every task verified by hand-written SQL, checking goal completion, state integrity and policy compliance🔥
𝐓𝐡𝐞 𝐡𝐞𝐚𝐝𝐥𝐢𝐧𝐞: Claude Opus 4.5 — our best-performing model succeeds on just 37.4% of tasks. With oracle tool access. No tool discovery required.
📄 arxiv.org/abs/2603.13594 (trending #4 on daily-papers)
🌐 enterpriseops-gym.github.io
🤗 huggingface.co/datasets/Serv…
💻 github.com/ServiceNow/Enterp…
Segan Subramanian retweeted
🚀 Introducing AprielGuard — an 8B parameter state-of-the-art guardian model.
Built to catch policy violations and jailbreaks alike, fully post-trained on synthetic data using SyGra.
🧠 Model Weights: huggingface.co/ServiceNow-AI…
📄 Blog: huggingface.co/blog/ServiceN…
@JayKasundra09 @SeganBoss @SathwikTejaswi @ServiceNow @ServiceNowNews @NVIDIAAI @nvidianewsroom
Why it matters 👇

ALT Bar chart titled ‘Safety Risks & Adversarial Attacks Performance’ showing mean F1 scores (higher is better) for safety models. AprilGuard-8B ranks highest at 90. IBM Granite-Guardian models follow: 3.3-3B (87), 3.1-2B (86), 3.2-5B (86), and 3.2-3B (82). Qwen3Guard-8B-Strict scores 73, tied with gpt-oss-safeguard-20B. Llama-guard-4-12B and Qwen3Guard-8B-Loose score 70. Llama-guard-3-8B scores 66, Llama-guard-2-8B scores 58, ShieldGemma-9B scores 52, and Llama-Prompt-Guard-2-86M is lowest at 44.
Segan Subramanian retweeted
Apriel-1.5-15B-Thinker sits in the “most attractive quadrant” of the Intelligence vs. Parameters tradeoff — alongside much larger proprietary models.
And among 4B–40B models, it ranks at the top.
The secret? 🤔









