DatabricksLanguage model
DBRX
DBRX has no overall index value yet because there are too few independent benchmark results.
PopularityRank 710 of 785
19out of 100-1 vs. last month
based on two data groups, mostly search
Summary
DBRX's defining feature is its fine-grained MoE architecture with 132 billion total parameters and 36 billion active. It selects four of 16 experts per token and was trained on roughly 12 trillion tokens. The instruct checkpoint combines a 32K context with coding, RAG, and SQL use for controlled enterprise deployment.
Databricks released base and instruct weights and described API use for Databricks customers alongside private deployment. This description concerns the instruct checkpoint, not DBRX Base.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 20 published benchmark results
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.
Open LLM Leaderboard aggregate (Show measurement, test conditions, and source)
- Source value
- 74.5
- Score
- 74.5
- Metric
- percent
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks DBRX release benchmark report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. DBRX Instruct's leaderboard average over ARC Challenge, HellaSwag, MMLU, TruthfulQA, WinoGrande, and GSM8K as described by Databricks; snapshot/protocol details not supplied.
Databricks Gauntlet v0.3 (Show measurement, test conditions, and source)
- Source value
- 66.8
- Score
- 66.8
- Metric
- percent
- Unit
- %
- Benchmark version
- v0.3
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks DBRX release benchmark report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Publisher-reported mean score on Databricks' internal Gauntlet v0.3 benchmark; task details and sample count are not published in the post.
HumanEval (Show measurement, test conditions, and source)
- Source value
- 70.1
- Score
- 70.1
- Metric
- pass@1
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks DBRX release benchmark report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. DBRX Instruct; Databricks compares against CodeLlama-70B-Instruct. The post does not specify prompt, harness, or decoding setup.
MMLU (Show measurement, test conditions, and source)
- Source value
- 73.7
- Score
- 73.7
- Metric
- percent
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks DBRX release benchmark report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. DBRX Instruct; the report table states 5-shot. Remaining prompt and decoding details are not stated.
HellaSwag (Show measurement, test conditions, and source)
- Source value
- 89
- Score
- 89
- Metric
- percent
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks DBRX release benchmark report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. DBRX Instruct; report table states 10-shot. Remaining prompt and decoding details are not stated.
GSM8K CoT majority@1 (Show measurement, test conditions, and source)
- Source value
- 72.8
- Score
- 72.8
- Metric
- majority@1
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks DBRX release benchmark report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. DBRX Instruct; chain-of-thought, report's updated harness result; table reports majority@1. This is distinct from the older 66.9 result.
DBRX Long-Context Table 3, Answer in Beginning Third of Context (Show measurement, test conditions, and source)
- Source value
- 45.1
- Score
- 45.1
- Metric
- Average performance across KV-Pairs and HotpotQAXL
- Unit
- %
- Benchmark version
- Databricks DBRX technical blog, Table 3, published 2024-03-27
- Category
- long-context question answering
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks AI Research Team, Introducing DBRX: A New State-of-the-Art Open LLM
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. The exact table column is DBRX Instruct. Table 3 averages performance over KV-Pairs (from Lost in the Middle) and HotpotQAXL (a longer-context modification of HotPotQA). This cell is the source row "Answer in Beginning Third of Context". The report says DBRX Instruct was trained with up to a 32K-token context window. It does not identify a serving/API revision, decoding setup, per-cell N or confidence interval.
- Context
- Vendor-reported average across KV-Pairs and HotpotQAXL, not a result on either benchmark in isolation. The report gives no per-cell denominator or confidence interval.
DBRX Long-Context Table 3, Answer in Middle Third of Context (Show measurement, test conditions, and source)
- Source value
- 45.3
- Score
- 45.3
- Metric
- Average performance across KV-Pairs and HotpotQAXL
- Unit
- %
- Benchmark version
- Databricks DBRX technical blog, Table 3, published 2024-03-27
- Category
- long-context question answering
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks AI Research Team, Introducing DBRX: A New State-of-the-Art Open LLM
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. The exact table column is DBRX Instruct. Table 3 averages performance over KV-Pairs (from Lost in the Middle) and HotpotQAXL (a longer-context modification of HotPotQA). This cell is the source row "Answer in Middle Third of Context". The report says DBRX Instruct was trained with up to a 32K-token context window. It does not identify a serving/API revision, decoding setup, per-cell N or confidence interval.
- Context
- Vendor-reported average across KV-Pairs and HotpotQAXL, not a result on either benchmark in isolation. The report gives no per-cell denominator or confidence interval.
DBRX Long-Context Table 3, Answer in Last Third of Context (Show measurement, test conditions, and source)
- Source value
- 48
- Score
- 48
- Metric
- Average performance across KV-Pairs and HotpotQAXL
- Unit
- %
- Benchmark version
- Databricks DBRX technical blog, Table 3, published 2024-03-27
- Category
- long-context question answering
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks AI Research Team, Introducing DBRX: A New State-of-the-Art Open LLM
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. The exact table column is DBRX Instruct. Table 3 averages performance over KV-Pairs (from Lost in the Middle) and HotpotQAXL (a longer-context modification of HotPotQA). This cell is the source row "Answer in Last Third of Context". The report says DBRX Instruct was trained with up to a 32K-token context window. It does not identify a serving/API revision, decoding setup, per-cell N or confidence interval.
- Context
- Vendor-reported average across KV-Pairs and HotpotQAXL, not a result on either benchmark in isolation. The report gives no per-cell denominator or confidence interval.
DBRX Long-Context Table 3, 2K Context (Show measurement, test conditions, and source)
- Source value
- 59.1
- Score
- 59.1
- Metric
- Average performance across KV-Pairs and HotpotQAXL
- Unit
- %
- Benchmark version
- Databricks DBRX technical blog, Table 3, published 2024-03-27
- Category
- long-context question answering
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks AI Research Team, Introducing DBRX: A New State-of-the-Art Open LLM
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. The exact table column is DBRX Instruct. Table 3 averages performance over KV-Pairs (from Lost in the Middle) and HotpotQAXL (a longer-context modification of HotPotQA). This cell is the source row "2K Context". The report says DBRX Instruct was trained with up to a 32K-token context window. It does not identify a serving/API revision, decoding setup, per-cell N or confidence interval.
- Context
- Vendor-reported average across KV-Pairs and HotpotQAXL, not a result on either benchmark in isolation. The report gives no per-cell denominator or confidence interval.
DBRX Long-Context Table 3, 4K Context (Show measurement, test conditions, and source)
- Source value
- 65.1
- Score
- 65.1
- Metric
- Average performance across KV-Pairs and HotpotQAXL
- Unit
- %
- Benchmark version
- Databricks DBRX technical blog, Table 3, published 2024-03-27
- Category
- long-context question answering
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks AI Research Team, Introducing DBRX: A New State-of-the-Art Open LLM
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. The exact table column is DBRX Instruct. Table 3 averages performance over KV-Pairs (from Lost in the Middle) and HotpotQAXL (a longer-context modification of HotPotQA). This cell is the source row "4K Context". The report says DBRX Instruct was trained with up to a 32K-token context window. It does not identify a serving/API revision, decoding setup, per-cell N or confidence interval.
- Context
- Vendor-reported average across KV-Pairs and HotpotQAXL, not a result on either benchmark in isolation. The report gives no per-cell denominator or confidence interval.
DBRX Long-Context Table 3, 8K Context (Show measurement, test conditions, and source)
- Source value
- 59.5
- Score
- 59.5
- Metric
- Average performance across KV-Pairs and HotpotQAXL
- Unit
- %
- Benchmark version
- Databricks DBRX technical blog, Table 3, published 2024-03-27
- Category
- long-context question answering
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Databricks AI Research Team, Introducing DBRX: A New State-of-the-Art Open LLM
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. The exact table column is DBRX Instruct. Table 3 averages performance over KV-Pairs (from Lost in the Middle) and HotpotQAXL (a longer-context modification of HotPotQA). This cell is the source row "8K Context". The report says DBRX Instruct was trained with up to a 32K-token context window. It does not identify a serving/API revision, decoding setup, per-cell N or confidence interval.
- Context
- Vendor-reported average across KV-Pairs and HotpotQAXL, not a result on either benchmark in isolation. The report gives no per-cell denominator or confidence interval.
Page 1 of 2
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
DBRXCodestral-22B-v0.1 19Llama 3.1 70B 19Step 3.5 Flash 2603 19Voxtral Small (latest) 19
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 DBRX, chosen by model line, developer, size, price and release date.
- WizardLM-2 8x22BMicrosoft · similar size
- Mixtral 8x22BMistral AI · similar size
- Mistral Large 2.1Mistral AI · similar size
- Pixtral Large (latest)Mistral AI · similar size
- Falcon 180BTII · similar size
- Code Llama 70B InstructMeta · similar size
- Qwen 3.8 Flash NextAlibaba · similar size
- Mixtral 8x7BMistral AI · similar size
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 DBRX
How large is the context window of DBRX?
32,768 tokens. That is how much text DBRX reads at once, about 50 A4 pages.
Page count as a rule of thumb from Artificial Analysis
GLM-5.1200K
DeepSeek-V3 0324164K
GLM-4.5131K
GPT OSS 120B131K
DeepSeek-V3.2128K
DeepSeek-R1128K
DBRXRank 103 of 10333K
Can I run DBRX locally?
Yes, the weights are open. With 132B parameters it needs a lot of memory, more a server than a laptop.
Stated by Databricks
- Open weights
- yes
- License
- Databricks Open Model
- Parameters
- 132B
- active per token
- 36B
Sources and data date
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. 2 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | benchmark-harness-and-snapshots Note The report omits exact harness versions, decoding settings, and dataset snapshots for most reported scores; older and updated GSM8K results must remain distinct. |
| Additional source | Databricks DBRX release benchmark report (retrieved October 15, 2026; October 18, 2026; October 21, 2026; October 22, 2026) · Databricks DBRX launch announcement · Editorial description reviewed October 2, 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 |