Thinkingmachines

Inkling

Generally availableOpen weights256K contextReasoning

Summary

Inkling is an open-weight language model from Thinkingmachines, released on 15 Jul 2026. It accepts text, images and PDFs and generates text, with a 256K-token context window and up to 256K output tokens per response. As of 18 Nov 2026, the lowest listed API price is $1 per million input tokens and $4.05 per million output tokens (Hugging Face Inference Providers), across 4 providers we track.

Multimodal MoE reasoning model (975B total, 41B active) for text, image, and audio

Last verified 18 Nov 2026. Every figure on this page links to its source.

Specifications

DeveloperThinkingmachines
Release date15 Jul 2026 [models.dev]
StatusGenerally available [models.dev]
WeightsOpen weights [Hugging Face]
Licenseapache-2.0 [Hugging Face]
Parameters952B [Hugging Face]
Context window256K (256,000 tokens) [models.dev]
Max output256K tokens [models.dev]
InputText, images and PDFs [models.dev]
OutputText [models.dev]
Reasoning modeYes [models.dev]
Tool callingYes [models.dev]
Structured outputYes [models.dev]
Hugging Facethinkingmachines/Inkling [Hugging Face]

Inkling API pricing by provider

USD per million tokens. Sorted by input price.

ProviderInputOutputCached inputContextSource
Hugging Face Inference Providers
thinkingmachines/Inkling
$1$4.05–1Mmodels.dev, 18 Nov 2026
Together AI
thinkingmachines/Inkling
$1$4.05$0.17524Kmodels.dev, 18 Nov 2026
OpenRouter
thinkingmachines/inkling
$1$4.05$0.17524KOpenRouter, 18 Nov 2026
Vercel AI Gateway
thinkingmachines/inkling
$1$4.05$0.17256Kmodels.dev, 18 Nov 2026

Running Inkling locally

Estimated memory, from the parameter count and the architecture in config.json on Hugging Face. Includes about 0.8 GB of runtime overhead.

QuantizationWeightsTotal, 8K contextTotal, 32K context
Q4_K_M (4-bit)577 GB580 GB587 GB
Q8_0 (8-bit)1012 GB1015 GB1022 GB
FP16 / BF16 (16-bit)1905 GB1908 GB1914 GB

At 4-bit with an 8K conversation, Inkling needs about 580 GB, more than any single consumer graphics card; it needs a unified-memory machine, a multi-GPU setup or a split between graphics card and system RAM. Check your own hardware, or see Best computer for local AI on our sister site AIPCs.

Frequently asked questions

What is the context window of Inkling?

Inkling has a context window of 256,000 tokens (256K) and can generate up to 256,000 tokens in a single response, according to models.dev.

How much does the Inkling API cost?

Inkling is listed from $1 per million input tokens and $4.05 per million output tokens (Hugging Face Inference Providers). Prices last verified 18 Nov 2026.

Is Inkling open source?

Inkling is an open-weight model: its weights are published on Hugging Face as thinkingmachines/Inkling under the apache-2.0 license. Check the license terms before commercial use.

When was Inkling released?

Thinkingmachines released Inkling on 15 Jul 2026, according to models.dev.

Which providers offer Inkling?

We track Inkling on 4 providers: Hugging Face Inference Providers, Together AI, OpenRouter, Vercel AI Gateway.

How much VRAM does Inkling need?

About 580 GB at 4-bit (Q4_K_M) with an 8K-token conversation, and about 1015 GB at 8-bit. That is more than any single consumer graphics card holds. These are estimates from the model's published parameter count and architecture.

Similar models from other labs

Comparable on output type, weights, release date, price and context window. Prices are the lowest listed, USD per million tokens.

ModelLabReleasedContextInput $/1MOutput $/1MWeights
Kimi K2.7 CodeMoonshot AI12 Jun 2026262K$0.60$3.50Open
Qwen3.8 2.4T A95BAlibaba (Qwen)12 Aug 2026262K$2$6Open
Kimi K2.6Moonshot AI20 Apr 2026262K$0.65$3.40Open
Qwen3.6 27BAlibaba (Qwen)22 Apr 2026262K$0.30$3.19Open
GlimpseElorian17 Oct 2026262K$0.60$3–
Aion 3.5Aion Labs23 Sept 2026262K$3$6–

Change history

  • Max output changed from 1M to 256K
  • Input modalities changed from text, image to text, image, pdf
  • Context window changed from 1M to 256K
  • Max output changed from 256K to 1M
  • Structured output changed from no to yes
  • Context window changed from 256K to 1M
  • Input modalities changed from text, image, pdf to text, image
  • Input modalities changed from text, image, audio to text, image, pdf
  • Context window changed from 524K to 256K
  • Max output changed from 472K to 256K
  • Input modalities changed from text, image to text, image, audio
  • Max output changed from 1M to 472K
  • Structured output changed from yes to no
  • Context window changed from 1M to 524K
  • Context window changed from 524K to 1M
  • Max output changed from 131K to 1M
  • Input modalities changed from text, image, audio to text, image
  • Input modalities changed from text, image, pdf to text, image, audio
  • Structured output changed from no to yes
  • Context window changed from 256K to 524K
  • Max output changed from 256K to 131K
  • Max output changed from 472K to 256K
  • Input modalities changed from text, image, audio to text, image, pdf
  • Context window changed from 524K to 256K
  • Input modalities changed from text, image, pdf to text, image, audio

Sources