Inkling-Small ranks #12 among open Agent Arena models

Inkling-Small’s strongest pitch is efficiency, not outright leaderboard dominance: a 12B-active model that preserves broad multimodal and agentic capability could be more useful than a marginally stronger, far costlier model.

• –Its 276B total and 12B active parameters make it substantially lighter to run than Inkling’s 975B total and 41B active configuration.
• –The reported $0.09 per task versus $0.20 for Inkling makes repeated coding and tool-use workloads materially more economical.
• –The negative net-improvement score suggests the model still has reliability or usability gaps in real agent workflows.
• –Native text, image, and audio reasoning plus a 1M-token context window give it a broader deployment surface than many small open models.
• –Developers should evaluate it on their own harnesses rather than treating the Agent Arena rank as a universal quality score.