Pick for RAG answering: Muse Spark 1.3 at $48.92 per month. Best cited long-context result for reading retrieved chunks: 1,048,576-token window, billed at $1.25 per 1M input on this workload, MRCR 98.3. Prices from Models.dev. Data last refreshed Oct 16, 2026, 07:15 AM UTC.
Job
Which model should read retrieved chunks in a RAG pipeline, and what will it cost each month?
This ranks the model that reads retrieved text, not the embedding model. Embeddings and rerankers are not in this catalog. The score weighs the context window, the input rate on a chunk-sized prompt, and a cited MRCR result. Cards are chosen only from models with that citation. A long window is not proof the model uses every chunk.
120 questions a day. Each prompt is about 12,000 tokens of retrieved chunks plus the question, and the answer is about 600 tokens. 30% of the input is treated as cached.
205 models cleared the bar.304 failed the lab, provider, or specialty bar.12 failed a capability filter.1 listed at $0 were left out.0 had no usable price.
Top pick
Muse Spark 1.3$48.92 / month
Best cited long-context result for reading retrieved chunks: 1,048,576-token window, billed at $1.25 per 1M input on this workload, MRCR 98.3.
Score 71.3 · $1.25 input · $4.25 output per 1M · context 1,048,576 · 16 providers
Budget pick
DeepSeek V4 Flash$5.23 / month
Lower bill with a cited MRCR result: 1,000,000-token window, billed at $0.14 per 1M input on this workload, MRCR 78.7.
Score 65.9 · $0.14 input · $0.28 output per 1M · context 1,000,000 · 52 providers
Premium pick
Qwen3.8 Max$76.68 / month
Stronger cited MRCR result among the models that cost more: 1,000,000-token window, billed at $2.00 per 1M input on this workload, MRCR 92.9.
Score 64.3 · $2.00 input · $6.00 output per 1M · context 1,000,000 · 32 providers
Skip this page when you are choosing an embedding or a reranker, or when the whole document has to sit in one prompt. Whole-document questions are on the long-document page, and that workload is much larger.
The reference workload is 120 questions a day, about 12,000 tokens in and 600 out, with a 30% cache hit. Change the calculator for a denser retrieval setup. The cards reprice; they do not re-rank.
The three cards follow the inputs. Fit, and who holds each role, stays on the reference workload.
| Caps | ||||||||
|---|---|---|---|---|---|---|---|---|
| 71.3 | Muse Spark 1.3 | $48.92 | $1.25 | $4.25 | 1,048,576 | 16 | Sep 2, 2026 | |
| 65.9 | DeepSeek V4 Flash | $5.23 | $0.14 | $0.28 | 1,000,000 | 52 | Apr 24, 2026 | |
| 65.3 | DeepSeek V4 Pro | $49.01 | $1.32 | $3.96 | 1,000,000 | 58 | Apr 24, 2026 | |
| 64.3 | Qwen3.8 Max | $76.68 | $2.00 | $6.00 | 1,000,000 | 32 | Aug 3, 2026 | |
| 57.9 | Muse Spark 1.2 | $48.92 | $1.25 | $4.25 | 1,048,576 | 18 | Aug 5, 2026 | |
| 20.4 | GPT-6 Luna | $4.23 | $0.10 | $0.50 | 1,050,000 | 34 | Sep 22, 2026 | |
| 18.8 | GPT-5.6 Luna | $8.90 | $0.20 | $1.20 | 1,050,000 | 46 | Jul 9, 2026 | |
| 16.3 | Llama 4 Scout 17B Instruct | $12.31 | $0.25 | $0.70 | 10,000,000 | 6 | Apr 5, 2025 | |
| 16.0 | Qwen3.7 Plus | $16.07 | $0.40 | $1.60 | 1,000,000 | 30 | Jun 2, 2026 | |
| 16.0 | Qwen3.5 Plus | $22.46 | $0.40 | $2.40 | 1,000,000 | 11 | Feb 16, 2026 | |
| 14.9 |  GLM-4.7-FlashX | $3.11 | $0.07 | $0.40 | 200,000 | 9 | Jan 19, 2026 | |
| 14.8 |  Nemotron 3 Ultra 550B A55B | $22.46 | $0.50 | $2.50 | 1,000,000 | 22 | Jun 4, 2026 |
