OpenAI text-embedding-3-large (2024年)（オープンエーアイテキストエンベディング3ラージ）
2024年OpenAI発表text-embedding-3-large・Industry-leading OpenAI flagship embedding model + Industry-leading 3072-dim Matryoshka + Industry-leading 8191 tokens + Industry-leading $0.13 per 1M tokens + Industry-leading API closed。
概要
OpenAI text-embedding-3-large は、2024年OpenAI発表text-embedding-3-large・Industry-leading OpenAI flagship embedding model 2024年 + Industry-leading API closed position確立。text-embedding-3-large specifications = Industry-leading OpenAI flagship embedding model (Industry-leading text-embedding-3-large 2024 + Industry-leading OpenAI flagship + Industry-leading text-embedding-3-large flagship) + Industry-leading 3072-dim Matryoshka (Industry-leading 3072-dim Matryoshka representation learning) + Industry-leading 8191 tokens (Industry-leading 8191 tokens max context) + Industry-leading $0.13 per 1M tokens (Industry-leading $0.13 USD per 1M tokens pricing)。
主な特徴・仕組み
- Org: Industry-leading OpenAI
- Year: 2024年 (2024年1月25日発表)
- Model: Industry-leading text-embedding-3-large (flagship) + text-embedding-3-small (cheaper)
- Embedding Dim: Industry-leading 3072-dim default (Matryoshka flexible 256-3072)
- Max Context: Industry-leading 8191 tokens max context
- Matryoshka: Industry-leading Matryoshka representation learning (truncation OK)
- MTEB Performance: Industry-leading MTEB 54.9% average score (large) / 62.3% (small)
- Multilingual: Industry-leading multilingual support 100+ languages
- Pricing: Industry-leading $0.13 USD per 1M tokens (large) / $0.02 (small)
- API Only: Industry-leading API closed only (no model weights released)
- Industry Adoption: Industry-leading widespread enterprise adoption ChatGPT API ecosystem
- Predecessor: Industry-leading text-embedding-ada-002 superseded
- Industry-Leading: Industry-leading OpenAI flagship embedding model 2024 Matryoshka
スペック比較表
| LLM Embedding Model (2023-2024) | Dim | Org | License | Industry Position |
|---|---|---|---|---|
| OpenAI text-embedding-3-large | 3072-dim Matryoshka | OpenAI | Closed API $0.13/1M | Industry-leading OpenAI closed API |
| Voyage-3 | 1024-dim | Voyage AI | Closed API $0.06/1M | Industry-leading Voyage closed API |
| BGE Large | 1024-dim BERT-large | BAAI | MIT open source | Industry-leading open-source MTEB top |
| E5-Mistral-7B | 4096-dim Mistral-7B | Microsoft | MIT open source | Industry-leading 7B Mistral embedding |
| Nomic Embed v1 | 768-dim Matryoshka | Nomic AI | Apache 2.0 fully open | Industry-leading fully open + reproducible |
具体例・対応製品
- OpenAI text-embedding-3-large (2024年): Industry-leading OpenAI flagship embedding
- Industry-leading 3072-dim Matryoshka flexible 256-3072: Industry-leading 3072-dim Matryoshka
- Industry-leading 8191 tokens max context: Industry-leading 8191 tokens
- Industry-leading $0.13 USD per 1M tokens pricing: Industry-leading $0.13/1M
- Industry-leading MTEB 54.9% large + 62.3% small: Industry-leading MTEB 54.9%+62.3%
- 競合 BGE + E5-Mistral + Nomic + Voyage: Industry-leading embedding competitors
自作PCでの選び方・注意点
OpenAI text-embedding-3-large は「Industry-leading OpenAI flagship embedding + Industry-leading 3072-dim Matryoshka」「Industry-leading 8191 tokens + Industry-leading $0.13 per 1M tokens + Industry-leading API closed」用途のIndustry-leading OpenAI flagship 2024年OpenAI発表product。Industry-leading 3072-dim Matryoshka (Industry-leading 3072-dim default Matryoshka flexible 256-3072 + Industry-leading Matryoshka representation learning + Industry-leading OpenAI Matryoshka signature) で Industry-leading 3072-dim Matryoshka + Industry-leading Matryoshka + Industry-leading OpenAI Matryoshka signature。Industry-leading 8191 tokens (Industry-leading 8191 tokens max context + Industry-leading text-embedding-3 long context + Industry-leading OpenAI long context signature) で Industry-leading 8191 tokens + Industry-leading long context + Industry-leading OpenAI signature。Industry-leading $0.13/1M (Industry-leading $0.13 USD per 1M tokens large + $0.02 small + Industry-leading OpenAI pricing tier + Industry-leading enterprise API priced) で Industry-leading $0.13/1M large + Industry-leading $0.02 small + Industry-leading enterprise API priced。Industry-leading MTEB 54.9% (Industry-leading MTEB 54.9% average score large + 62.3% small + Industry-leading OpenAI MTEB benchmark + Industry-leading text-embedding-3 performance) で Industry-leading MTEB 54.9% + Industry-leading 62.3% small + Industry-leading text-embedding-3 performance。Industry-leading multilingual + closed API (Industry-leading multilingual support 100+ languages + Industry-leading API closed only no weights + Industry-leading ChatGPT API ecosystem adoption + Industry-leading widespread enterprise adoption) で Industry-leading 100+ languages + Industry-leading API closed + Industry-leading ChatGPT ecosystem + Industry-leading enterprise widespread。Industry-leading ada-002 superseded (Industry-leading text-embedding-ada-002 superseded + Industry-leading OpenAI 2024 generation + Industry-leading text-embedding-3-large + text-embedding-3-small dual tier) で Industry-leading ada-002 superseded + Industry-leading 2024 generation + Industry-leading large+small dual tier。但しIndustry-leading BGE + E5-Mistral + Nomic + Voyage competition (Industry-leading BGE Large BAAI 1024-dim MIT open 2023 + E5-Mistral Microsoft 4096-dim Mistral-7B MIT 2023 + Nomic Embed v1 768-dim Apache 2024 + Voyage-3 1024-dim closed $0.06/1M 2024 vs OpenAI text-embedding-3-large 3072-dim Matryoshka 8191 closed $0.13/1M 2024 trade-off) で Industry-leading 4-embedding model competitors vs OpenAI text-embedding-3-large + Industry-leading 3072-dim Matryoshka + 8191 tokens + $0.13/1M + MTEB 54.9% large+62.3% small + multilingual 100+ languages + ChatGPT API ecosystem + closed API only enterprise adoption unique advantage adoption alignment必須。
関連用語との違い
- vs Voyage-3 (Closed $0.06/1M 2024): OpenAIは3072-dim + Matryoshka + $0.13・Voyageは1024-dim + retrieval-optimized + $0.06 cheaper
- vs BGE/E5-Mistral/Nomic (Open-source): OpenAIはclosed API + Matryoshka 3072 + 100+ languages・Open-sourceはMIT/Apache + weights public
- vs ada-002 (predecessor): text-embedding-3-largeはMatryoshka + 3072-dim + MTEB 54.9%・ada-002は1536-dim only
よくある質問（FAQ）
Q1: OpenAI text-embedding-3-large vs Voyage-3 違いは？ A: OpenAI text-embedding-3-large (Industry-leading 3072-dim default Matryoshka flexible 256-3072 + 8191 tokens max context + Matryoshka representation learning truncation OK + MTEB 54.9% large + multilingual 100+ languages + $0.13 USD per 1M tokens large + ChatGPT API ecosystem + OpenAI 2024) vs Voyage-3 (Industry-leading 1024-dim + 32K tokens max context + retrieval-optimized + Voyage AI MIT alum founders + $0.06 USD per 1M tokens cheaper + Voyage AI 2024)・Industry-leading 3072-dim + Matryoshka + 100+ languages + ChatGPT + $0.13 = OpenAI + Industry-leading 1024-dim + retrieval-optimized + cheaper $0.06 + Voyage = Voyage preference judgment。
Q2: Industry-leading 3072-dim Matryoshka + 8191 tokens value は？ A: Industry-leading 3072-dim + 8191 tokens (Industry-leading 3072-dim default Matryoshka flexible 256-3072 + Industry-leading Matryoshka representation learning truncation OK + Industry-leading 8191 tokens max context + Industry-leading OpenAI Matryoshka + long context signature)。
Q3: Industry-leading $0.13/1M + ChatGPT API ecosystem value は？ A: Industry-leading $0.13/1M + ChatGPT ecosystem (Industry-leading $0.13 USD per 1M tokens large + Industry-leading $0.02 small + Industry-leading OpenAI pricing tier + Industry-leading API closed only + Industry-leading ChatGPT API ecosystem adoption + Industry-leading widespread enterprise adoption)。
まとめ
OpenAI text-embedding-3-large = 2024年OpenAI発表のOpenAI flagship embedding model。Industry-leading 3072-dim default Matryoshka representation learning flexible 256-3072 + Industry-leading Matryoshka truncation OK + Industry-leading 8191 tokens max context + Industry-leading MTEB 54.9% average score large + 62.3% small + Industry-leading multilingual support 100+ languages + Industry-leading $0.13 USD per 1M tokens large + $0.02 small pricing + Industry-leading API closed only no model weights released + Industry-leading text-embedding-3-large + text-embedding-3-small dual tier + Industry-leading text-embedding-ada-002 superseded + Industry-leading ChatGPT API ecosystem adoption + Industry-leading widespread enterprise adoption + Industry-leading OpenAI flagship embedding model 2024 Matryoshka position確立。