Sulat.com
AI models
$10 off the fastest DeepSeek V4.1 Flash, Kimi K3 and GLM 5.3 from Synthetic
Helicone logo

Model details

Meta Llama Prompt Guard 2 86M

Meta Llama Prompt Guard the cataloged API limit 86M is a compact language model rooted in the established Llama lineage, designed for high-performance text processing with an emphasis on efficiency and scalability. Its architecture is engineered to handle focused language inputs with precision, making it particularly suitable for developers who need robust language capabilities without the overhead of larger models. The "Prompt Guard" designation signals its role as a safeguard-oriented tool, likely optimized for analyzing and filtering text inputs in production pipelines where rapid execution and resource management are critical.

Positioned within the Llama ecosystem, this model benefits from the research foundation that has made the family widely adopted across language tasks. It offers developers a streamlined solution for integrating language capabilities into workflows that demand speed, consistency, and cost-effective processing. Practical use cases include high-volume text generation, rapid input analysis, and infrastructure optimization where the balance between capability and computational footprint matters most. As part of a broader ecosystem of language models, it provides a reliable option for projects requiring consistent performance across diverse text-based applications.

Heliconellama-prompt-guard-2-86mllama

Quick Info

Powered by
Provider
Helicone
Model key
llama-prompt-guard-2-86m
Release date
Oct 1, 2024
Last updated
Oct 1, 2024
Knowledge cutoff
2024-10
Input modalities
Output modalities
Capabilities

Cost

Input token cost
$0.01
Output token cost
$0.01

Limits

Output tokens
2 tokens
Context window
512 tokens

Latest news about Meta Llama Prompt Guard 2 86M

Helicone

CoverageBenchmark

A February 2026 preprint titled "When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift" (arXiv:2602.14161, by Max Fomin and co-authors, surfaced via ResearchGate) systematically benchmarks Meta's PromptGuard 2 and LlamaGuard against 18 datasets covering harmful requests, jailbreaks, The study finds that PromptGuard 2, LlamaGuard, and LLM-as-judge approaches all fail on indirect attacks targeting agents, achieving only 7-37% detection, and that PromptGuard 2 and LlamaGuard cannot evaluate agentic tool injection at all due to architectural limitations (no chat-template support, and strict user/assis

Videos about Meta Llama Prompt Guard 2 86M

More models around Meta Llama Prompt Guard 2 86M