Multi-step reasoning
Security and privacy
Turn acquisition and asset data into faster investment decisions.
Models built for the work that run where your data lives
Reasoning, coding and clinical models, post-trained by DataObserve and available in DO products or through the DO Console API.
// Leap today withDataObserve
Our models
Each model is tuned for one kind of work, measured on public benchmarks and deployable in your own environment.
- // Reasoning & documents
DO Mek
Reasoning and document model for conversation, logic and connected workflows.
Works through multi-step problems, reads long PDFs, scans and tables with every claim grounded to a source page, and calls tools to finish tasks. Powers Coworker.
- Multi-step reasoning
- Document AI
- Multilingual
- Tool use
Benchmarks- 89.8
- GPQA Diamond
- 92.0
- OmniDocBench
- 94.9
- MathVision
- Parameters
- 27B
- Context
- 262K tokens, extensible to ~1M
- Input / output
- Text, image and video in; text out
- Access
- Coworker · DO Console API
Internal evaluations against the Qwen3.8-27B base; scores vary by harness and prompt.
- // Agentic coding
DO Mini
Intelligence for long-running agents and everyday development.
Edits across files, runs terminal workflows and stays on goal over hundreds of tool calls, with fewer tokens, steps and cost per task. Powers Developer.
- Multi-file edits
- Terminal workflows
- Long-horizon persistence
- Effort calibration
Benchmarks- 63.1
- SWE-bench Pro
- 74.6
- Terminal-Bench 2.1
- 90.8
- LiveCodeBench v6
- Parameters
- 27B
- Context
- 262K tokens, extensible to ~1M
- Max output
- 65,536 tokens
- Access
- Developer · DO Console API
Internal evaluations against the Qwen3.8-27B base; scores vary by harness and prompt.
- // Clinical
DO Med
Medical foundation model for imaging, clinical text and de-identification.
Structured symptom assessments, findings linked to the source image region or text, and PHI removal across all 18 HIPAA Safe Harbor identifiers, for practitioner review.
- Medical imaging
- Anatomical localization
- Biomedical entities
- PHI de-identification
Benchmarks- 87.7
- MedQA
- 74.2
- MedMCQA
- 87.0
- MMLU Med
- Architecture
- Decoder-only transformer (GQA)
- Context
- 128K tokens
- Input / output
- Text and imaging in; JSON / FHIR out
- Deployment
- In your environment; no patient data leaves
Benchmark results, not evidence of clinical safety or clearance. Every output needs clinician review.
// Model Catalog
Open-weight models, optimized fortraining and inference
Every model below is open-weight, optimized for training and inference, and tested for correctness. For on-prem model and hardware requirements
- Qwen3.8 · DenseQwen3.8 27B27B
- Qwen3.6 · MoEQwen3.6 35B-A3B35B-A3B
- Qwen3.5 · MoEQwen3.5 397B-A17B397B-A17B
- Kimi · MoEKimi K2.6Open weights
- Kimi · MoEKimi K3Open weights
- DeepSeek · MoEDeepSeek V4 ProOpen weights
- Gemma 4 · MoEGemma 4 26B-A4B26B-A4B
- gpt-oss · MoEgpt-oss 120B120B
- GLM · MoEGLM 5.3Open weights
- GLM · MoEGLM 5.3 FlashOpen weights
- DeepSeek · MoEDeepSeek V4 FlashOpen weights
- Explore all 174 models