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
Inkling 2 Small is a 594B sparse MoE with 17B active parameters, native vision and a 1M-token context. Run it in the Tinker Playground today.
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.
Veo3.1
Create Stunning Videos with Veo3.1
FLUX 3 Video Generator
Kling 3.0
Next-Gen AI Video Generator
Grok Video Generator
Create Videos from Text or Images with AI
MiniMax H3 video generator
MiniMax H3 AI Video Generator
Why Inkling 2 Small Stands Out
A next-gen sparse mixture-of-experts system that reads images natively, scales effort per request and holds up to 1M tokens of context.
- 594B Parameters, Only 17B ActiveEach token travels through a 48-layer decoder-only transformer that activates 8 of 384 experts, mixing local and global attention.
- Tunable Effort, Million-Token ContextSet reasoning effort per call, push up to 1M tokens, and upload images sized between 40px and 4096px on each side.
- Launching Closed, Opening Under Apache 2.0It debuts on Tinker behind closed weights; the full checkpoints follow under Apache 2.0 after outside safety testing and red-teaming wrap up.
Three Ways to Run Inkling 2 Small
Pick the path that fits you on Tinker — try it, tune it or download it.
What Inkling 2 Small Can Do
Extended autonomy, design work that sees its own output, and steady reasoning across code, research and knowledge work.
Agentic Coding Over Long Horizons
Reinforcement learning across hundreds of thousands of extended coding tasks in many agent harnesses and compaction setups, scored on method, not just green tests.
GPU Kernels and ML Autoresearch
It optimizes GPU kernels on PMPP-Hard and trained Qwen3.5-4B on Tinker in 4.5 hours for $426, more than doubling that model's forecasting score.
Vision Encoder and Grounding Stack
A 292M Perception Encoder front end using 20×20-pixel patches with 2×2 spatial pooling, solid on OCR, grounding, infographics and visual reasoning.
Checking Its Own Rendered Output
It inspects what it renders and iterates on websites, games and 3D scenes, and identified three wines from one photo in 19 of 20 attempts.
Knowledge Work and Finance Agents
MCP tool calls, a 1527 ELO on AA-Briefcase, and the best open-weights numbers on multiple finance benchmarks covering filings and spreadsheets.
Alignment Plus Two-Layer Monitoring
CBRN results sit below the open-weight frontier with a lower FORTRESS Average Risk Score, backed by a synchronous pre-execution tool monitor and an async LLM judge.
Inkling 2 Small — Common Questions
Quick answers on size, weights, inputs, training and benchmarks.
How big is Inkling 2 Small?
It carries 594B total parameters while activating 17B per token, sending each one to 8 of 384 experts in a 48-layer decoder-only transformer.
When will the open weights land?
At launch they remain closed on Tinker; the full checkpoints are slated for an Apache 2.0 drop the following week, once outside safety testing and red-teaming are complete.
Which inputs and context length are supported?
UTF-8 text plus pixel images, ideally 40px to 4096px on each side, returning text and accepting a context window of up to 1M tokens.
What did the training run look like?
8.4T pretraining 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 a joint RL run of 1,075 steps with 17M+ rollouts across 9 days.
Where does it top the benchmarks?
It posts 31 on the Artificial Analysis Intelligence Index, 1527 ELO on AA-Briefcase, the best open-weights finance numbers, and a DesignArena Website top score still pending.
Can I adapt it to my own data?
Yes — on identical data and recipes it picks things up faster than earlier models, matching Inkling Small's final score on a financial-document summarization task within 75 training steps.
Run Inkling 2 Small in the Tinker Playground Now
Head into the Tinker Playground for instant inference, kick off tuning with the Tinker Cookbook, and be ready for the Apache 2.0 weight drop.