Language modelActive
DeepSeek-R1-Distill-Llama-70B
DeepSeek
- Released
- January 20, 2025
- Data date
- September 8, 2026
- Speed
- 21.7 tokens/sOutput throughput
- Price
- Historical rates under Costs
- Input
- Text
- Output
- Text
What is distinctive about DeepSeek-R1-Distill-Llama-70B is the transfer of R1 reasoning patterns into a dense 70B checkpoint based on Llama 3.3 70B. The smaller local checkpoint does not inherit the mixture-of-experts architecture of DeepSeek-R1; it distills the original reasoning data instead.
DeepSeek-R1-Distill-Llama-70B is technically and legally separate from R1. Redistribution requires checking both the DeepSeek terms and the Llama license.
Specifications and access
| Specification | Value and source |
|---|---|
| Weights | Open weightsSource |
| Access | LocalSource |
| Parameters | 70BSource |
| Context window | 128,000 tokensSource |
| License | MITSource |
| Parameters | 70 BSource |
| Architecture | Llama 3.3 distilled reasoning modelSource |
| Context window | 128,000 tokens (Context inherited from the R1 card; exact distilled tokenizer context not separately stated)Source |
| Reasoning mode | distilled reasoning modelSource |
Page 1 of 2
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
AIME24 (Show measurement, test conditions, and source)
- Source value
- 70
- Score
- 70
- Metric
- percent
- Unit
- %
- Category
- reasoning
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek R1 model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. distill comparison table; source states Llama-3.3-70B base
AIME24 cons@64 (Show measurement, test conditions, and source)
- Source value
- 86.7
- Score
- 86.7
- Metric
- percent
- Unit
- %
- Category
- reasoning
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek R1 model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. distill comparison table; source states Llama-3.3-70B base
MATH500 (Show measurement, test conditions, and source)
- Source value
- 94.5
- Score
- 94.5
- Metric
- percent
- Unit
- %
- Category
- reasoning
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek R1 model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. distill comparison table; source states Llama-3.3-70B base
GPQA (Show measurement, test conditions, and source)
- Source value
- 65.2
- Score
- 65.2
- Metric
- percent
- Unit
- %
- Category
- reasoning
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek R1 model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. distill comparison table; source states Llama-3.3-70B base
LiveCode (Show measurement, test conditions, and source)
- Source value
- 57.5
- Score
- 57.5
- Metric
- percent
- Unit
- %
- Category
- reasoning
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek R1 model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. distill comparison table; source states Llama-3.3-70B base
Codeforces rating (Show measurement, test conditions, and source)
- Source value
- 1633
- Score
- 1633
- Metric
- rating
- Unit
- rating
- Category
- reasoning
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek R1 model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. distill comparison table; source states Llama-3.3-70B base
Artificial Analysis Intelligence Index (Show measurement, test conditions, and source)
- Source value
- 7.88824581191087
- Score
- 7.88824581191087
- Metric
- points
- Unit
- points
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. unknown
AA-Omniscience Index (Show measurement, test conditions, and source)
- Source value
- -46.3833333333333
- Score
- -46.3833333333333
- Metric
- points
- Unit
- points
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. unknown
Output speed (Show measurement, test conditions, and source)
- Source value
- 21.6836262610367
- Score
- 21.6836262610367
- Metric
- tokens/s
- Unit
- tokens/s
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. unknown; default 10K-input workload only
- Context
- Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: serving-provider median. Deprecated model. Artificial Analysis only continues benchmarking the default 10K-input workload.
Latency reasoning time (Show measurement, test conditions, and source)
- Source value
- 92.23549492705467
- Score
- 92.23549492705467
- Metric
- seconds
- Unit
- seconds
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. unknown; default 10K-input workload only
- Context
- Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: serving-provider median. Deprecated model. Artificial Analysis only continues benchmarking the default 10K-input workload.
Latency input time (Show measurement, test conditions, and source)
- Source value
- 0.97468479600002
- Score
- 0.97468479600002
- Metric
- seconds
- Unit
- seconds
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. unknown; default 10K-input workload only
- Context
- Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: serving-provider median. Deprecated model. Artificial Analysis only continues benchmarking the default 10K-input workload.
Answer generation time (end-to-end timing breakdown) (Show measurement, test conditions, and source)
- Source value
- 23.05887373176367
- Score
- 23.05887373176367
- Metric
- seconds
- Unit
- seconds
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. unknown; default 10K-input workload only
- Context
- Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: serving-provider median. Deprecated model. Artificial Analysis only continues benchmarking the default 10K-input workload.
Page 1 of 2
Popularity
33.2 / 100
Sources and data date
| Type | Evidence and data date |
|---|---|
| Speed | Median across serving providers. Output speed after the first streaming chunk, not total response time. Classified as deprecated by Artificial Analysis. Artificial Analysis maintains only the default 10K-input workload there. Retrieved September 8, 2026. |
| Input | Documented modalities. Retrieved September 29, 2026. |
| Output | Documented modalities. Retrieved September 29, 2026. |
| Research status | Research date September 8, 2026. 3 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Unresolved | license Original wording Distilled weights combine DeepSeek release terms with Llama 3.3 license; the checked card does not provide a single unambiguous effective license field. |
| Unresolved | release-date Original wording No normalized release-date field. |
| Additional source | Context window data · DeepSeek-R1 release and distillation details · Editorial description reviewed October 2, 2026 |
| Additional source | API pricing |
| Additional source | DeepSeek R1 model card (retrieved September 8, 2026) |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Creative Commons Attribution 4.0 |