128k+
Best LLM for enterprise RAG
Models that tick the three boxes of an internal RAG: long context (128k tokens and up), controllable hosting (self-hosted or EU option) and enough capability to synthesise without inventing. Ranked by Elo.
| # | Model | Arena Elo | Context | Hosting | Licence |
|---|---|---|---|---|---|
| 1 | 🇺🇸Claude Opus 4.7 Anthropic | 1502 | 500K | EU option | proprietary |
| 2 | 🇺🇸Gemini 3.7 Flash Google | 1491 | 1M | EU option | proprietary |
| 3 | 🇺🇸Gemini 3 Pro Google | 1486 | 2M | EU option | proprietary |
| 4 | 🇺🇸Claude Opus 4.8 Anthropic | 1482 | 1M | EU option | proprietary |
| 5 | 🇺🇸Gemini 3 Flash Google | 1474 | 1M | EU option | proprietary |
| 6 | 🇺🇸Claude Sonnet 4.6 Anthropic | 1472 | 200K | EU option | proprietary |
| 7 | 🇨🇳Kimi K2.6 Moonshot AI | 1461 | 262K | Self-hostable | modified-mit |
| 8 | 🇨🇳GLM-5 Zhipu AI | 1458 | 128K | Self-hostable | apache-2.0 |
| 9 | 🇺🇸Gemma 4 31B Google | 1451 | 256K | Self-hostable | apache-2.0 |
| 10 | 🇨🇳Qwen3.5 397B-A17B Alibaba | 1441 | 262K | Self-hostable | apache-2.0 |
Data verified on 4 September 2026Open in the comparator
Method
Deterministic ranking on comparator data: LMArena Elo (style control) first, HuggingFace downloads second for models without an Elo. Prices and licences are verified at the official source and dated. No model is ranked by an LLM.