50 models7 providersprices scraped nightly — no estimates
databricks · dbrx family

DBRX Instruct

132B params · 33K context · databricks-open-model · released 2024-03

DBRX Instruct is a mixture-of-experts model from Databricks released March 2024, with 132B total parameters and roughly 36B active per token across a 16-expert architecture. The finer-grained expert count — 16 small experts versus Mixtral's 8 larger ones — was a deliberate design choice to improve routing granularity; at launch it outperformed Mixtral 8x7B on MMLU, HumanEval, and several reasoning benchmarks. Context window is 32K. The Databricks Open Model License is permissive for most commercial use but is not OSI-approved, so legal review is warranted before deployment. Since mid-2024 it has been largely overtaken by DeepSeek V3 and Mixtral 8x22B on cost-adjusted quality, so the main case for DBRX today is teams already embedded in the Databricks ecosystem.

Cheapest now
no hosted pricing yet
90-day price Δ
cheapest-host trend
Fastest TTFT
no data
Providers
0
indexed providers

No pricing data available for this model yet.

No hosted pricing scraped for this model yet. Methodology · Raw data