None
None
Long Story Video Skill
Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.
Ads Video Skill
Generate professional ads and sales videos—AI auto-generates scripts, prompts, and footage.
3D Science Explainer Video Skill
Convert scientific concepts into stunning 3D explain animations
Feedback
freeTrialImage.bannerPity
freeTrialImage.upgradeUnlock
- ✓freeTrialImage.benefitHd
- ✓freeTrialImage.benefitWatermark
- ✓freeTrialImage.benefitUnlimited
Inkling 2 Small
Explore Inkling 2 Small — a 594B-parameter sparse model activating 17B per token, with native image input and a 1M-token window. Try it in Tinker.
All Tools
Discover our comprehensive AI-powered animation toolkit
MiniMax H3
MiniMax H3 AI Video Generator
Seedance 2.5
The Future of AI Video Is Here.
Seedance 2.0
The Future of AI Video Is Here.
Seedance 2.0
The Future of AI Video Is Here.
Gemini Omni
Gemini Omni Video Generator
Image to Video
Generate videos from images
Gemini Omni 1.2 Flash
AI Multi-Scene Shorts Generator
Create viral AI Shorts instantly
Core Strengths of the Inkling 2 Small Model
A next-wave sparse Mixture-of-Experts design that takes pictures directly, scales effort per call, and holds a million tokens in view.
- 594B Parameters, Only 17B Fired per TokenEach token travels through 8 of 384 experts inside a 48-layer decoder-only transformer that blends local and global attention.
- Tunable Reasoning Depth, 1M-Token ReachSet how hard the model thinks on each call, push in up to a million tokens, and pass images between 40px and 4096px per side.
- Closed Weights First, Apache 2.0 LaterIt debuts on Tinker with weights held back; the full set should land under Apache 2.0 once outside safety testing and red-teaming wrap up.
Getting Started With Inkling 2 Small
Three practical routes into the model, all on Tinker.
Capabilities Built Into Inkling 2 Small
Extended autonomy, design loops that keep eyes on the output, and well-calibrated reasoning transfer across code, research and knowledge work.
Agentic Coding Over Long Horizons
Reinforcement learning across hundreds of thousands of extended coding jobs in varied harnesses and compaction setups, scored on method as well as passing tests.
Self-Directed Kernel and ML Research
Agent-driven GPU kernel tuning on PMPP-Hard, plus one experiment that spent 4.5 hours and $426 on Tinker training Qwen3.5-4B to more than double its forecasting score.
Dedicated Vision Encoder With Grounding
A 292M-parameter Perception Encoder stack using 20×20-pixel patches and 2×2 spatial pooling, holding up on OCR, grounding, infographics and visual reasoning.
Checking Its Own Work Visually
The model inspects its own rendered sites, games and 3D scenes and iterates, and it named three wines from a single photo in 19 of 20 trials.
Knowledge Work and Finance Agents
MCP tool calling, an ELO of 1527 on AA-Briefcase, and top open-weights numbers on several finance benchmarks covering filings and spreadsheets.
Safety Alignment With Layered Oversight
CBRN results sit under the open-weight frontier with a lower Average Risk Score on FORTRESS, backed by a synchronous pre-execution tool monitor and an asynchronous LLM judge.
Inkling 2 Small: Questions Answered
Straight answers on the size, weights, inputs, training and benchmark standing of Inkling 2 Small.
How big is the Inkling 2 Small model?
It holds 594B parameters in total while activating 17B per token, sending each one to 8 of 384 experts within a 48-layer decoder-only transformer.
When will the open weights be released?
At launch the weights stay closed on Tinker; the full set is slated to ship under Apache 2.0 the following week, once outside safety testing and red-teaming are complete.
Which inputs and how much context does it handle?
UTF-8 text plus pixel images, best between 40px and 4096px per side, returning text and accepting a context window of as much as 1M tokens.
What did the training run look like?
Pretraining covered 8.4T tokens over 68 hours on NVIDIA GB300 NVL72 racks in native NVFP4, then a 3.8T-token stage at 512K context, 320B reasoning and agent tokens at 1M context, 34B high-quality agent trajectories, and finally a joint RL run of 1,075 steps and more than 17 million rollouts across 9 days.
Where does it top the benchmarks?
It posts 31 on the Artificial Analysis Intelligence Index, ELO 1527 on AA-Briefcase, leading finance results among open weights, and a top DesignArena Website score still pending.
Can I adapt it to my own data?
Yes. On the same data and recipe it picks things up faster than earlier models, matching Inkling Small's final score on a financial-document summarization task in only 75 training steps.
Run Inkling 2 Small in the Tinker Playground Now
Head into the playground for instant inference, kick off fine-tuning from the Tinker Cookbook, and keep an eye out for the Apache 2.0 weights.