Span-01 model built to monitor and evaluate other AI
Evaluators and safety teams typically pay for slow, expensive GPT calls or accept brittle classifiers; Span-01 aims to replace both with one cheap judgment pass. Teams running LLM guardrails, red-teaming pipelines or production monitoring can plug it in as a low-cost scorer, starting with the free Lite tier. At US$0.02 per million input tokens, it undercuts most general-purpose API scoring setups.
Span-01 is a model designed specifically to monitor and evaluate the outputs of other AI models.
It combines reasoning and judgment in a single pass and outputs a probability directly, so it can check multiple rules at once while reading long conversation transcripts.
It handles brand-new judgment rules written on the fly by developers, which traditional classifiers fail at, while running faster and cheaper than calling large models like GPT for scoring.
In tests, Span-01 scored above comparable specialised small models and some mainstream large models at detecting jailbreak attacks, privacy leaks, hallucinations and tool-calling errors.
Pricing includes a free Span-01 Lite version and a full version at US$0.02 per million input tokens.
Span-01：专门用来监测和评估其他AI的模型
质检专用版Jev。
以往监测 AI 是否犯错，很多团队习惯直接调用 GPT 这样的大型语言模型来逐字打分，或者使用传统的分类程序。传统的分类程序运行速度快，但面对未见过的全新问题时容易失误；大型通用模型虽然判断准确，但逐字输出的机制导致运行速度慢，花费也很高。
Span-01 解决了这个问题。它把推理和判定合在了一次运算中完成，直接输出结果概率。就算面对开发人员临时给出的全新判定规则，Span-01 也能一边看懂长对话记录，一边同时核对多条规则。
在针对 AI 实际表现的测试中，Span-01 在发现越狱攻击、隐私泄露、胡言乱语以及工具调用错误等问题上的表现，综合得分超过了同类的专用小模型，也超过了部分主流大型模型。
费用方面，官方推出了完全免费的 Span-01 Lite 版本，而完整版 Span-01 每百万输入词元的调用费用为 0.02 美元。
怎么一下子冒出好多这种模型？
官方介绍：https://t.co/K6Y0Jc3b3u
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison and AI regulation, compliance frameworks, and sovereign compute policy. Key entities and technical topics under continuous tracking include Span-01, Jev, AI safety, monitoring.
Source attribution: curated from Gorden Sun (@Gorden_Sun).
