Inkling

[reasoning]

Open-weight 975B MoE (41B active) with native reasoning, multimodal understanding, and fine-tuning via the Tinker platform. Apache 2.0 licensed.

Input

per 1M tokens

Output

$4.05

per 1M tokens

Context

1.0M tokens

Max Output

66K tokens

Released

2026-07-15

Features

[optional-reasoning][multimodal-input][tool-calling][open-weights][fine-tunable][long-context-(1m)]

Technical Specifications

LicenseApache 2.0

Architecture

ArchitectureSparse Mixture of Experts (MoE)

Parameters975B total / 41B active

Resources

Paper HuggingFace Demo

Benchmarks

AIME 202697.1%

Artificial Analysis Intelligence Index41

GPQA Diamond87.2%

Humanity's Last Exam29.7%

IFBench79.8%

MCP Atlas74.1%

SWE-bench Verified77.6%

SimpleQA Verified43.9%

Terminal-Bench63.8%

The Lab

location: San Francisco, California, USAfounded: 2025backed by: Andreessen Horowitz, NVIDIA; $2B raised

Founded in 2025 in San Francisco by former OpenAI CTO Mira Murati, Thinking Machines Lab has raised $2 billion from backers including Andreessen Horowitz and NVIDIA. The lab focuses on frontier model training and building customisable, steerable AI platforms, positioning itself as an independent challenger to the OpenAI/Anthropic/Google triad.

more about Thinking Machines Lab →

Model ID: inkling