Inkling 2 Small Online Demo
Drop in a prompt or picture, then collect step-by-step logic, working code and ready-to-ship drafts
None

None

None

Long Story Video Skill

Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.

Ads Video Skill

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

AI Video Prompt Generator

Feedback

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

  • ✓freeTrialImage.benefitHd
  • ✓freeTrialImage.benefitWatermark
  • ✓freeTrialImage.benefitUnlimited

Inkling 2 Small

Test Inkling 2 Small free in your browser: it reads images, reasons step by step and writes code — no install, no credit card, instant results.

All Tools

Discover our comprehensive AI-powered animation toolkit

What You Get With Inkling 2 Small

A second-generation sparse Mixture-of-Experts design that sees images natively, scales reasoning effort per request and holds an enormous working context.

  • Sparse Experts: 594B Total, 17B Active
    Every token is routed to 8 of 384 experts across 48 decoder-only layers that blend local and global attention for speed.
  • Tunable Thinking, Vast Context
    Raise or lower reasoning effort per call, feed very long inputs, and pass images from 40px to 4096px a side.
  • Hosted First, Open Weights to Follow
    It debuts on Tinker behind gated weights; the full release lands under Apache 2.0 once outside safety review and red-teaming wrap up.

Three Ways to Start With Inkling 2 Small

Pick the path that fits your workflow — nothing to install up front.

What Inkling 2 Small Can Do

Autonomy across extended tasks, design work that checks its own renders, and steady reasoning that carries over into code, research and knowledge jobs.

Coding Agents That Stay on Task

Reinforcement learning over huge volumes of extended coding jobs and agent setups, judged on the quality of the approach rather than green tests alone.

GPU Kernels and ML Autoresearch

Agent-driven kernel tuning on PMPP-Hard, plus a 4.5-hour, $426 Tinker run that more than doubled Qwen3.5-4B's forecasting score.

A Dedicated Vision Encoder

A 292M-parameter Perception Encoder using 20×20-pixel patches and 2×2 spatial pooling, holding up on OCR, grounding, infographics and visual reasoning.

It Checks Its Own Renders

The model inspects what it produced and iterates on sites, games and 3D scenes — and named three wines from one photo in 19 of 20 tries.

Knowledge Work and Finance Agents

MCP tool calls, an ELO of 1527 on AA-Briefcase, and the best open-weights scores on several finance benchmarks covering filings and spreadsheets.

Safety Guardrails in Two Layers

CBRN results under the open-weight frontier, a lower FORTRESS Average Risk Score, a pre-execution tool monitor and an asynchronous LLM judge.

FAQ

Inkling 2 Small: Your Questions Answered

Quick answers on model size, inputs, training and open-weight timing.

1

How big is the model?

594B parameters in total, with 17B switched on per token through 8 of 384 experts inside a 48-layer decoder-only transformer.

2

When will the weights be released openly?

They stay gated on Tinker at launch. A full Apache 2.0 release is planned for the following week, once external safety testing and red-teaming wrap up.

3

What can I feed it?

UTF-8 text plus pixel images — ideally 40px to 4096px per side — with text output and a working window of up to 1M tokens.

4

How was it trained?

8.4T pretraining tokens in 68 hours on GB300 NVL72 racks in native NVFP4, a 3.8T-token stage at 512K context, 320B reasoning and agent tokens at 1M, 34B top agent trajectories, then a joint RL run of 1,075 steps and 17M+ rollouts over 9 days.

5

Where does it score best?

31 on the Artificial Analysis Intelligence Index, ELO 1527 on AA-Briefcase, top open-weights finance results, and a DesignArena Website score still pending.

6

Can I adapt it to my own data?

Yes. It picks up new material faster than earlier models — matching Inkling Small's final score on financial-document summarization in only 75 training steps.

Start Building With Inkling 2 Small

Fire up the playground to test the model right now, spin up a fine-tune from the Cookbook, and keep an eye out for the Apache 2.0 weights.