Inkling
Thinking Machines · released Jul 14, 2026 · thinkingmachines/Inkling
- Type
- Open weightsApache License 2.0
- Params
- 952B
- Context
- 1M
about 786K words of context
Our take
Written Sep 17, 2026Inkling is a downloadable model from Thinking Machines that takes text, images and audio in the same request, and its licence allows commercial use, changes and redistribution. The catch is scale: at 952.4 billion parameters it is a hosted proposition rather than something you run on a desktop.
Reach for it when a single request has to carry text, a screenshot and a recording together, or when the work is mathematical and analytical and the measured maths results are the ones you care about. Its licence allows commercial use, changes and redistribution, so it can sit inside a product. Skip it if you meant to run the model on your own machine, or if agentic coding is the job.
The case for it
- 88.36% on LiveBench Mathematics, an average over competition-style tasks rather than over real project work, so treat it as a signal for analytical jobs and check it on your own.
- Text, image and audio all go into one request, so a recording or a screenshot does not have to be transcribed or described before it goes in.
- The request capacity holds a long report or a stack of documents beside the question, though reliable recall across all of it is unverified in our data.
The case against it
- Agentic coding is its weakest measured area: 49.39% on LiveBench Agentic Coding inside an agent harness, against 71.02% on the standalone coding tasks, so weigh the harness figure for agent work.
- 952.4 billion parameters in total with no active-parameter figure supplied, so a hosted offer is the practical route rather than a machine of your own.
- Its arena results are mixed: creative writing and web-app building sit at the lower end of its own preference results, while coding and maths prompts sit higher.
How good is it?
An open text model for chat and everyday questions, though multi-step tasks and changes of direction are where it struggles.
- carrying out multi-step tasks for youArena Agent · 47th of 55
- changing course when you give new instructionsArena Agent · Steerability · 53rd of 55
EverydayGeneral questions and everyday reasoning
Arena Text (overall)56th of 168 · 1442
CodingWriting and fixing code on its own
Arena Coding59th of 168 · 1491
AgenticPlanning, calling tools, staying on task
Arena Agent47th of 55 · −0.108
WritingDrafting and rewriting prose
Arena Creative Writing83rd of 168 · 1384
Placings on Arena's agent boards, from live sessions people ran themselves. A model can lead on one of these and sit mid-field on the others.
Boards this model appears on that none of the ratings above are built on.
Every published score for this model20 scoresEvery figure we hold, from 20 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
GeForce RTX 4090 · 24 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Pro (16-core GPU) · 32 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Radeon RX 7900 XT · 20 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked 4 hours ago — each listing carries its own date.
- per 1M tokens
- $1.00 in / $4.05 out
- Context served
- 524K
- Throughput
- ~75 tok/s
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| OpenRouterOpenRouter's own listing | $1.00 / $4.05checked 4 hours ago | 524K | not measured | Unknown | Unknown | Unknown |
| DeepInfrafp8Direct and through OpenRouter | $0.95 / $4.05checked 4 hours ago | 524K262K max reply through OpenRouter | 48 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| Together AIThrough OpenRouter | $1.00 / $4.05checked 4 hours ago | 524K472K max reply | 75 tok/s | No | No | Confirmed |
OpenRouter
DeepInfraAcross the 3 listings we hold: 2 say they do not train on prompts (1 of them only through OpenRouter), 0 say they do and 1 does not say. 2 appear in the zero-retention registry we check (1 of them only through OpenRouter); the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| OpenRouterOpenRouter's own listing | ✓ | ✗ | ✗ |
| DeepInfrafp8Direct and through OpenRouter | ✓ | ✗ | ✗ |
| Together AIThrough OpenRouter | ✓ | ✗ | ✗ |
Tool calling: 3 of 3 listings say yes. JSON output: 0 of 3 listings say yes, 3 say no. Strict schema: 0 of 3 listings say yes, 3 say no.
When we formed this view
Recent changes
Each date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 1 of 3 listings does not say whether it trains on prompts, and 1 answers only through OpenRouter, not for its own listing.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
Licence and identifiers
What the licence allowsApache License 2.0, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Apache License 2.0
Fully permissive: commercial use, redistribution, and derivatives allowed. Requires attribution and a copy of the license. Includes an express patent grant.
Identifiers
- Hugging Face
- thinkingmachines/Inkling
- Architecture
- Dense
- Takes in, gives back
- Text, images and audio in, text out
- Catalogue slug
- thinkingmachines-inkling