Inkling 2 Small goes open, hits AI Gateway

Inkling 2 Small is a serious open-weight contender, but its “small” label hides infrastructure requirements that put local deployment out of reach for most developers.

• –17B active parameters enable efficient inference relative to its 594B total size
• –RL on long-horizon coding tasks targets real agent loops, not just isolated code-generation benchmarks
• –AI Gateway adds provider routing, failover, and compatibility with AI SDK, OpenAI, and Anthropic APIs
• –Multimodal inputs and a 1M-token context window broaden its usefulness beyond coding
• –Native NVFP4 checkpoints require roughly 500GB of aggregated VRAM, making hosted inference the practical default