← Back to Docs
Gemma 4 26B
Google's open-weight model, optimized for spec generation and structured reasoning.
| Provider | Google DeepMind |
| Parameters | 26 billion |
| Context | 128K tokens |
| Best for | Spec generation, structured output, planning |
| Dispatch role | spec_agent, test_spec_agent |
| Infrastructure | Dispatch Pro cloud |
| Cost | Included in Pro ($20/month, shared 5M token pool) |
Overview
Gemma 4 26B excels at generating well-structured specifications from natural language goal descriptions. Its strong instruction-following capabilities make it particularly effective as a spec agent in Dispatch's pipeline, where it produces detailed, actionable specifications that downstream implementation agents can follow.
When Dispatch uses Gemma 4
- Spec generation -- When auto-model is enabled, Dispatch may route spec generation to Gemma 4 for goals that require structured planning and decomposition.
- Test spec generation -- Gemma 4 produces thorough test plans that cover edge cases and integration points.
- Reasoning tasks -- Any pipeline stage that benefits from structured reasoning can be routed to Gemma 4 via pipeline profile configuration.
Configuration
Gemma 4 is available to all Dispatch Pro subscribers. To use it:
- Open Settings > Pipeline Profiles
- Set any stage's agent to
gemma-4-26b - Or enable Auto Model to let Dispatch choose the best model per task
Isolation
All cloud model runs use worktree isolation. Dispatch creates an isolated copy of your repository for each cloud agent run, preventing any unintended changes to your working tree. Results are reviewed and merged through the standard HITL flow.