ModelPriceWatch.com
Last scan 2026-09-15 Models tracked 259 Providers 32 Cheapest paid Granite 4.0 H Micro $0.017/Mtok in Every price links to its source

Inkling

by Thinking Machines · 975B total / 41B active parameters

Current mid tier open weights mid tier
Email me when Inkling gets cheaper

Today's price · per 1M tokens

Input

$1.00

per 1M tokens

Output

$4.05

per 1M tokens

Blended

$1.76

blended $/1M — 3:1 weighted input:output

Cached input

$0.170

17% of input — prompt caching

How this price is scoped: Tinker Serverless Inference (Beta) list price for the 256K-context nvfp4 sampling endpoint, the only per-token inference rate Thinking Machines publishes. Tinker's separate training/fine-tuning tiers are priced per 1M tokens as well but bill prefill, sample and train legs individually and are not a hosted-inference list price.

Source: official Thinking Machines pricing · read MODELPRICEWATCH.COM · 2026-09-15

Price receipt

We read Thinking Machines's own pricing page on and found Inkling listed with a price on the same row — that read is where the number above comes from. We keep the page text we read; its content hash is 1e34d6686b.

Overview

Thinking Machines' first frontier model: a 975B-total / 41B-active Mixture-of-Experts transformer that reasons natively over text, images and audio, with controllable thinking effort. The weights are open on Hugging Face, and the lab also sells it directly through Tinker's serverless inference at $1.00/$4.05 per 1M tokens, with cached input at $0.17.

Run it yourself

Deploy this open model on rented GPUs

Open weights mean you can self-host instead of paying per-token API prices. These platforms let you serve it on demand — often cheaper at scale.

RunPod →
Rent GPUs by the second and serve any open-weight model yourself.
Novita AI →
Serverless open-model inference billed per token — no GPU ops.
Vast.ai →
GPU marketplace — rent spot/on-demand GPUs cheaply and serve any open-weight model.

Partner links — we may earn a commission if you sign up, at no cost to you. We only list platforms we'd recommend regardless.

Capabilities

struck through = not supported
Input 3/5
Text ✓ Image ✓ Audio ✓ Video PDF
Output 1/5
Text ✓ Image Audio Video Embedding
Features 5/9
Prompt caching ✓ Reasoning ✓ Coding Fast inference Long context ✓ Open weights ✓ Multimodal ✓ Web search Realtime

Benchmark performance

accuracy % · higher is better
Percentile vs all tracked models 43.3th4 independent measurements
Percentile per $/Mtok 24.6
GPQA Diamond vendor
87.2%
SWE-Bench Verified vendor
77.6%
Terminal-Bench 2.1 vendor
63.8%
MCP Atlas vendor
76%
BrowseComp vendor
77.1%
Humanity's Last Exam vendor
29.7%

Every per-benchmark score we hold for this model is reported by its own vendor, so none is counted toward an accuracy average or any ranking. The percentile above is its standing across the independent composites below.

Independent composite scores
  • AA Intelligence Index: 25.5 (Artificial Analysis composite across reasoning, knowledge and coding evals)
  • AA-Omniscience Index: 2 (−100–100)
  • GDPval-AA v2: 1165.2 ELO (Real-world work tasks, human baseline = 1000)
  • LMSYS Chatbot Arena: 1440.1 ELO (Human preference)
How it stacks up
  • Ranks #90 of 181 comparably-measured models by percentile score, across 4 independent measurements
  • Ranks #117 of 181 comparably-tested models by normalized performance per dollar

51.1 tokens/sec output

Source: Artificial Analysis, LMSYS Chatbot Arena (UC Berkeley), Thinking Machines launch post (Introducing Inkling) · updated · See full rankings →

Specifications

Provider
Thinking Machines
Context window
262K tokens
Modality
text, image, audio
Parameters
975B total / 41B active
Open source
Yes — open weights available
Released
Status
Current
Last updated
Tags
open-weightsmid-tierreasoningmultimodalmoecaching

Availability verified: — listed on Thinking Machines's own page

Official pricing page