Gemma 4 31B IT (Runware)

Gemma 4 31B IT (Runware) is a 31-billion-parameter instruction-tuned member of Google's Gemma family, distributed with open weights and surfaced through Runware as part of the LLM Gateway catalog. Its design targets reasoning-heavy agentic workloads: chain-of-thought tasks, complex logic, and tool-integrated workflows, with native function calling, adjustable temperature, and attachment support listed among the model fields. The same listing describes it as a cost-effective option at the gateway's published token rates, positioning the variant as a practical middle-ground between smaller open models and the larger proprietary frontier systems rather than as a research-only release.

The variant exposes text and image inputs with text-only outputs, making it usable for visual question answering and document-grounded chats while keeping responses strictly textual. It operates inside a 262K-token context window with a 66K-token output ceiling, which is generous enough for long retrieval-augmented sessions, multi-turn agent traces, and code repositories. Open weights, combined with reasoning capability and tool use, make the model a flexible base for self-hosted customization, evaluation pipelines, and integration into multi-model ensembles where smaller Gemma variants can hand off routine tasks to a more capable 31B reasoning core.