Breaking

Thinking Machines Lab's Open-Weight Inkling Ranks 9th on Design Arena's Agentic Web App Leaderboard

July 15, 2026 at 14:16 EDT

Foundation Models
Open Source
AI Agents

Inkling, the newly released open-weight model from Thinking Machines Lab, has placed 9th overall on Design Arena's Agentic Web App Arena with an Elo rating of 1257, putting it in the same performance band as leading proprietary systems from Anthropic and Google DeepMind.

Open-Weight AI · Thinking Machines Lab

Inkling Cracks the Top 10 on the Agentic Web App Arena

A newly released open-weight model lands 9th overall with an Elo of 1257 — sitting in the same performance band as leading proprietary systems from Anthropic and Google DeepMind.

overall on the Agentic Web App Arena

1257

Elo rating, top U.S. open-weight model on agentic work

5.1M+

voters across 190+ countries behind the score

Where Inkling stands on the leaderboard

Elo rating · Web Dev (Agentic) · taller column = higher rank

1344

Claude
Sonnet 5

1290

Claude
Opus 4.7

1262

Gemini
3.5 Flash

1257

Inkling
(open-weight)

The lone open-weight entry sits within a narrow band of top proprietary systems — a category closed models have long dominated.

Inside the model

975B

total parameters (Mixture-of-Experts)

41B

active per token · 12B in Small variant

45T

multimodal training tokens (text · image · audio · video)

token context window

Why it matters

Open weights are published for download, fine-tuning, and local execution — with a customization pathway through the Tinker platform, targeting organizations that want frontier-adjacent performance without closed APIs.

The caveat

The Elo is a subjective, preference-based measure — not task completion. Independent reviews of code quality and failure modes are still scarce, so durability depends on more votes and broader testing.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.