1stTop pick
Researched 15d agoNemotron 3 Super-120B-A12B
- RULER (128K)
- 96.79%
- Output (from)
- $0.400 / 1M
Best LLMs for Retrieval-Augmented Generation (2026)
As of 2026-10-26, LLM Reference's top pick for RAG is Nemotron 3 Super-120B-A12B by NVIDIA AI (RULER (128K) 96.79%, $0.400/1M output tokens), chosen from models that meet the podium eligibility rules. Second pick: MiniMax-01 by MiniMax.
Last refreshed 2026-10-26. Next refresh: weekly.
Best LLMs for RAG in 2026 — ranked by RULER score at 128K tokens. Other models without a 128K RULER score are not listed.
Verdict
Use Nemotron 3 Super-120B-A12B for RAG today.
MiniMax-01 is the runner-up, 2 points back on RULER (128K).
2ndShortlist
Researched 15d agoMiniMax-01
- RULER (128K)
- 94.7%
- Output (from)
- $1.10 / 1M
3rdShortlist
Researched 15d agoQwen3-235B-A22B-Instruct-2507
- RULER (128K)
- 93.9%
- Output (from)
- $0.550 / 1M
How we rank
RAG picks rank on one score: RULER at 128K tokens. Other retrieval benchmarks (ZeroSCROLLS, InfiniteBench, multi-needle, TriviaQA, Natural Questions, SQuAD, MS MARCO) are shown for reference and do not set the order.
- Eligibility — Only models with a RULER score at 128K tokens, ranked by that score.
- Primary ranking — Models with a RULER score at 128K tokens are ranked by that one score. Other retrieval benchmarks are shown for reference and do not set the order.
- Variant collapse — We keep one row per model family (`familySlug` + parameter tier). When headline scores tie within ±0.5 pt (±10 Elo on Chatbot Arena), we pick the canonical SKU by lowest tracked input price, then GA over preview or limited access, then newest `release`. A folded sibling within the benchmark noise band can show a "Tied within margin" chip on that score cell.
- Podium eligibility — Preview, restricted, retired and retiring models never fill a podium slot. Preview means preview or invite-only access, or a model that is announced or pre-release. Restricted means restricted, partner, private or otherwise unavailable access. Retired means the model is past its retirement date or every provider route is retired. Retiring means the model, or any one of its provider routes, has a retirement date within the next 60 days. These models can still appear in the table below the podium.
- Pricing — Lowest tracked provider input/output where present.
| # | Model | Input $/1M | Output $/1M | |
|---|---|---|---|---|
| 1 | Nemotron 3 Super-120B-A12B ReasoningTools RULER (128K): 96.79% | $0.09 | $0.40 | |
| 2 | MiniMax-01 Vision RULER (128K): 94.7% | $0.20 | $1.10 | |
| 3 | Qwen3-235B-A22B-Instruct-2507 RULER (128K): 93.9% | $0.09 | $0.55 | |
| 4 | Nemotron 3 Nano Tools RULER (128K): 91% | — | — | |
| 5 | Qwen3-235B-A22B RULER (128K): 90.6% | — | — | |
| 6 | Jamba 1.5 Mini ToolsVerify pricing RULER (128K): 90% | $0.20 | $0.40 | |
| 7 | Qwen3-30B-A3B-Instruct-2507 RULER (128K): 89.1% | $0.09 | $0.30 | |
| 8 | Qwen3-32B RULER (128K): 85.6% | $0.08 | $0.28 | |
| 9 | Qwen3-14B RULER (128K): 85.1% | $0.10 | $0.22 | |
| 10 | Qwen3-30B-A3B RULER (128K): 79.2% | $0.05 | $0.30 | |
| 11 | Nemotron-Nano-9B-v2 RULER (128K): 78.9% | $0.06 | $0.23 | |
| 12 | Qwen3-8B RULER (128K): 77.4% | $0.04 | $0.14 | |
| 13 | Yi 34B 200K Verify pricing RULER (128K): 77.3% | $0.20 | $0.90 | |
| 14 | Llama 3.1 8B Instruct Verify pricing RULER (128K): 77% | $0.02 | $0.05 | |
| 15 | Gemma 3 27B PT RULER (128K): 72.9% | $0.23 | $0.38 | |
| 16 | Llama 3 70B Gradient 1048K Verify pricing RULER (128K): 72.1% | — | — | |
| 17 | Llama 3 8B Gradient 1048K Verify pricing RULER (128K): 69.5% | — | — | |
| 18 | Phi-3 Mini 128K Verify pricing RULER (128K): 66.7% | $0.05 | $0.10 | |
| 19 | Llama 3.1 70B Instruct RULER (128K): 66.6% | $0.40 | $0.40 | |
| 20 | Qwen3-4B Verify pricing RULER (128K): 66% | $0.20 | $0.20 |
Honorable mentions
Next seats in this ranking. Lines below are from each model's stored description in LLMReference seed data—spot-check the model page before relying on a capability claim.
NVIDIA's lightweight 3.97B parameter model optimized for edge deployment with FP8 quantization (W8A8 mixed precision). Designed for agentic AI applications including gaming NPCs, local voice assistants, and IoT automation. Supports instruction following, tool use, and hallucination avoidance. Strong performance on BFCL, IFBench, IFEval, HaluEval, RULER, Tau2, AIME25, MATH500, GPQA-D, and LiveCodeBench.
91%
RULER (128K)
Qwen3-235B-A22B is Alibaba's Qwen3 model. It offers a 128K-token context window.
90.6%
RULER (128K)
Jamba 1.5 Mini is AI21 Labs' Jamba 1.5 model. It offers a 256K-token context window.
90%
RULER (128K)
Compare Top Picks
Side-by-side comparison of the top picks by price, benchmark, and API access.
Browse Other Categories
Best LLMs for Code GenerationBest LLMs for Agents vs Agent Products (2026)Best LLMs for ClassificationBest Open Source LLMsBest Multimodal / Vision LLMsBest LLM for Translation in 2026Best AI Image Models in 2026Best AI Video Models in 2026Best LLMs for Reasoning & MathBest Small Language Models (SLMs)Best LLMs for JSON / Tool useCheapest LLM APIs You Can Call Right NowBest Long Context LLMsBest Mainstream LLM APIs, RankedBest LLMs for EnterpriseBest Free LLMs You Can Use Right NowBest LLMs for WritingBest LLMs for MarketingBest LLMs for Customer Support
Frequently asked questions
What is the best LLM for RAG?
LLM Reference's top pick for RAG as of 2026-10-26 is Nemotron 3 Super-120B-A12B by NVIDIA AI, with RULER (128K) 96.79%. Picks come from models that meet the podium eligibility rules, so the top pick may not lead the full table. Second pick: MiniMax-01 by MiniMax (RULER (128K) 94.7%).
Which ranked LLM for RAG has the lowest output price?
Among the ranked models shown on this page, Llama 3.1 8B Instruct by AI at Meta has the lowest listed output price: $0.050/1M output tokens, at rank 14.
What is the best open-weight LLM for RAG?
Nemotron 3 Super-120B-A12B by NVIDIA AI is the highest-ranked open-weight model shown on this page (rank 1, RULER (128K) 96.79%). License: NVIDIA Nemotron Open Model License.
How is the RAG ranking calculated?
RAG picks rank on one score: RULER at 128K tokens. Other retrieval benchmarks (ZeroSCROLLS, InfiniteBench, multi-needle, TriviaQA, Natural Questions, SQuAD, MS MARCO) are shown for reference and do not set the order. Preview, restricted, retired and retiring models never fill a podium slot. Preview means preview or invite-only access, or a model that is announced or pre-release. Restricted means restricted, partner, private or otherwise unavailable access. Retired means the model is past its retirement date or every provider route is retired. Retiring means the model, or any one of its provider routes, has a retirement date within the next 60 days. These models can still appear in the table below the podium.
How often is this page updated?
Weekly. Last refreshed 2026-10-26.