Pronouncing Open Source LLM Names

Mistral is pronounced 'MIS-trul' — named after the powerful Mediterranean wind: MIS (as in 'miss'), then 'trul' (TRUL, schwa ending). Stress on MIS. Don't say 'mis-TRAL' with end stress. In a technical interview: "We use MIS-trul Small for most tasks and MIS-trul Large when we need more reasoning capacity — the cost difference is significant."

Phi is pronounced 'FY' — the Greek letter phi, rhymes with 'sky', 'by', and 'fly'. Single syllable. Don't say 'FEE' (that's the Greek letter pi alternative). In a technical interview: "We run Microsoft's FY-4 model on edge devices — it's a small language model that punches well above its weight class for reasoning tasks."

Gemma is pronounced 'JEM-uh' — the G is soft like in 'gem': JEM (rhymes with 'them'), then 'uh' (schwa). Stress on JEM. Don't say 'GEM-uh' with a hard G. In a technical interview: "We run JEM-uh 2 locally for privacy-sensitive summarization tasks — it's Google's lightweight open model that fits on a consumer GPU."

Qwen is pronounced 'CHWEN' — the Q in Chinese phonetics (pinyin) represents a CH sound, so it's CHWEN (like 'chwen' rhyming with 'when'). Don't say 'KWEN'. In a technical interview: "We evaluated CHWEN 2.5 from Alibaba — it's competitive with the best open models and supports over 100 languages."

Llama is pronounced 'LAH-muh' — like the animal: the double-L is a single L sound, LAH (long A as in 'father'), then 'muh' (schwa). Stress on LAH. Don't say 'LAM-uh' with a short A. In a technical interview: "We fine-tune LAH-muh 3 on domain-specific data using LoRA and serve it on our own inference cluster to keep costs predictable."