DeepSeekLanguage model
DeepSeek LLM 67B
DeepSeek LLM 67B has no overall index value yet because there are too few independent benchmark results.
25out of 100same as last month
based on search, usage and attention
Summary
DeepSeek LLM 67B Chat comes from DeepSeek’s 2023 LLM generation. Its 67 billion dense parameters and 4,096-token context window clearly separate it from the later V3 and R1 models.
The model card documents local text chat but neither multimodal input nor tool loops. The short context window remains this checkpoint’s central limitation.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 54 published benchmark results
Page 1 of 5
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.
GSM8K CoT (Show measurement, test conditions, and source)
- Source value
- 84.1
- Score
- 84.1
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek 67B Chat; Chain-of-Thought
- Category
- math
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek LLM author evaluation appendix
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact DeepSeek LLM 67B Chat model row/widget in DeepSeek LLM author evaluation appendix. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
MATH CoT (Show measurement, test conditions, and source)
- Source value
- 32.6
- Score
- 32.6
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek 67B Chat; Chain-of-Thought
- Category
- math
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek LLM author evaluation appendix
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact DeepSeek LLM 67B Chat model row/widget in DeepSeek LLM author evaluation appendix. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
MGSM Chinese CoT (Show measurement, test conditions, and source)
- Source value
- 74
- Score
- 74
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek 67B Chat; Chain-of-Thought
- Category
- math
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek LLM author evaluation appendix
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact DeepSeek LLM 67B Chat model row/widget in DeepSeek LLM author evaluation appendix. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
CMATH CoT (Show measurement, test conditions, and source)
- Source value
- 80.3
- Score
- 80.3
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek 67B Chat; Chain-of-Thought
- Category
- math
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek LLM author evaluation appendix
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact DeepSeek LLM 67B Chat model row/widget in DeepSeek LLM author evaluation appendix. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
Gaokao Math Cloze CoT (Show measurement, test conditions, and source)
- Source value
- 16.9
- Score
- 16.9
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek 67B Chat; Chain-of-Thought
- Category
- math
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek LLM author evaluation appendix
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact DeepSeek LLM 67B Chat model row/widget in DeepSeek LLM author evaluation appendix. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
Gaokao Math QA CoT (Show measurement, test conditions, and source)
- Source value
- 20.2
- Score
- 20.2
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek 67B Chat; Chain-of-Thought
- Category
- math
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek LLM author evaluation appendix
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact DeepSeek LLM 67B Chat model row/widget in DeepSeek LLM author evaluation appendix. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
GSM8K tool-integrated reasoning (Show measurement, test conditions, and source)
- Source value
- 86.7
- Score
- 86.7
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek 67B Chat; tool-integrated reasoning
- Category
- math
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek LLM author evaluation appendix
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact DeepSeek LLM 67B Chat model row/widget in DeepSeek LLM author evaluation appendix. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
MATH tool-integrated reasoning (Show measurement, test conditions, and source)
- Source value
- 51.1
- Score
- 51.1
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek 67B Chat; tool-integrated reasoning
- Category
- math
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek LLM author evaluation appendix
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact DeepSeek LLM 67B Chat model row/widget in DeepSeek LLM author evaluation appendix. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
HellaSwag (Show measurement, test conditions, and source)
- Source value
- 75.7
- Score
- 75.7
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek LLM report Table 6 (arXiv:2401.02954)
- Category
- English
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek AI, technical report
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. DeepSeek LLM 67B Chat author run, report Table 6, task shot count is not stated for this chat-model row.
- Context
- The paper reports the Table 6 value under its internal evaluation framework. Per-model scored N and CI are not reported.
PIQA (Show measurement, test conditions, and source)
- Source value
- 82.6
- Score
- 82.6
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek LLM report Table 6 (arXiv:2401.02954)
- Category
- English
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek AI, technical report
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. DeepSeek LLM 67B Chat author run, report Table 6, task shot count is not stated for this chat-model row.
- Context
- The paper reports the Table 6 value under its internal evaluation framework. Per-model scored N and CI are not reported.
WinoGrande (Show measurement, test conditions, and source)
- Source value
- 76
- Score
- 76
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek LLM report Table 6 (arXiv:2401.02954)
- Category
- English
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek AI, technical report
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. DeepSeek LLM 67B Chat author run, report Table 6, task shot count is not stated for this chat-model row.
- Context
- The paper reports the Table 6 value under its internal evaluation framework. Per-model scored N and CI are not reported.
RACE-Middle (Show measurement, test conditions, and source)
- Source value
- 70.9
- Score
- 70.9
- Metric
- accuracy
- Unit
- %
- Benchmark version
- DeepSeek LLM report Table 6 (arXiv:2401.02954)
- Category
- English
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- DeepSeek AI, technical report
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. DeepSeek LLM 67B Chat author run, report Table 6, task shot count is not stated for this chat-model row.
- Context
- The paper reports the Table 6 value under its internal evaluation framework. Per-model scored N and CI are not reported.
Page 1 of 5
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
The 100 most popular models of this category, 0 to 100, higher is more popular
20 of 101 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 7, 2026
DeepSeek LLM 67BIndex-Homura-9B 25Trinity Nano Preview 25ERNIE 4.5 VL 424B A47B 25Index-Homura-2B 25
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as DeepSeek LLM 67B, chosen by model line, developer, size, price and release date.
- DeepSeek-R1-Distill-Llama-70BDeepSeek · same developer
- DeepSeek-R1-Distill-Qwen-32BDeepSeek · same developer
- DeepSeek Coder V2DeepSeek · same developer
- DeepSeek-R1DeepSeek · same developer
- DeepSeek-V3DeepSeek · same developer
- DeepSeek R1 0528 Qwen3 8BDeepSeek · same developer
- DeepSeek-V3 0324DeepSeek · same developer
- DeepSeek V4.1 Flash 1017DeepSeek · same developer
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Frequently asked questions about DeepSeek LLM 67B
How large is the context window of DeepSeek LLM 67B?
4,096 tokens. That is how much text DeepSeek LLM 67B reads at once, about 10 A4 pages.
Page count as a rule of thumb from Artificial Analysis
Can I run DeepSeek LLM 67B locally?
Yes, the weights are open.
Stated by DeepSeek
- Open weights
- yes
- License
- DeepSeek License
- Parameters
- 67B
- active per token
- n/a
Sources and data date
Show all 10 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. |
| Output | Documented modalities. |
| Research status | Research date October 15, 2026. 3 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | benchmark-task-dates-and-non-math-units Note Math appendix explicitly labels percentages and distinguishes CoT from tool-integrated reasoning. Other table rows do not state run dates, scoring protocol, or units. |
| Not published | Table 6, AlignBench, MT-Bench and held-out per-model denominators and uncertainty Note The checked author tables do not report per-row scored N or CI. No sampleCount is inferred from benchmark corpus size or prompt count. |
| Unresolved | IFEval paper and repository value discrepancy Note arXiv Table 9 reports 55.5 for the prompt-level loose metric; the official evaluation/more_results.md reports 59.1 for Prompt-level Instruction Following. No run metadata reconciles them. |
| Additional source | DeepSeek LLM author evaluation appendix (retrieved October 15, 2026) |
| Additional source | DeepSeek AI, technical report (retrieved October 17, 2026) |
| Additional source | Model metadata (retrieved October 17, 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 | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |