Typhoon2 70B Instruct

Typhoon2 70B Instruct, provided by scb10x, is a 70-billion parameter Thai-English instruction-tuned language model built on Llama 3.1, created on March 28, 2025. It consistently performs among the fastest models, ranking in the Infinityth percentile across four benchmarks, indicating exceptional speed. Price data is unavailable, suggesting potential free tier usage. The model demonstrates exceptional reliability, achieving the 100th percentile across four benchmarks, meaning it consistently provides usable responses with minimal technical failures. Despite its strong foundational capabilities and high reliability, the model exhibited 0.0% accuracy across all evaluated baseline benchmarks, including Email Classification, Reasoning, Ethics, and General Knowledge. This indicates a significant weakness in accurately performing these specific tasks, despite its reported strengths in general instruction-following, math, coding, tool-use, and state-of-the-art results in Thai-specific benchmarks like IFEval and MT-Bench. Its core strengths lie in bilingual reasoning and function-calling scenarios, with comparative evaluations showing improvements over prior Thai LLMs and other Llama-based baselines, as detailed in its technical report. The current benchmark results suggest a need for further evaluation or fine-tuning for general English-centric baseline tasks.