SPARK Atlas 1
A Software Spark-developed model line for traceable operational reasoning, structured decisions, and tool-mediated workflows.
VERSION1.1FAMILYReasoning and tool useCONTEXTUp to 256K target; confirmed per deploymentRELEASEControlled preview
- INPUTS
- Text, structured context, tool definitions
- OUTPUTS
- Text, JSON schema, tool calls
- OWNERSHIP
- Software Spark-developed model program; artifact access subject to evaluation agreement.
- Operational decision support with human approval gates
- Structured extraction and classification
- Evidence-linked enterprise assistants
- Tool-use workflows with restricted action policies
- Not a substitute for qualified professional judgment
- May generate incorrect or incomplete conclusions
- Requires domain-specific evaluation and access controls
- Context and throughput depend on the deployed artifact and hardware
- Task-specific correctness and refusal criteria
- Structured-output and tool-call validity
- Evidence coverage and citation fidelity
- Latency and throughput using disclosed deployment hardware
No external benchmark score is claimed on this card. Reproducible results will follow the SPARK benchmark protocol.