Mira Murati’s Inkling AI Model Review: Best Open-Source Model in the West

Mira Murati spent two years constructing one thing new after leaving OpenAI, lastly revealing it to the general public final week.

Inkling, the first model from Murati’s Thinking Machines Lab, can be the most effective open-source mannequin skilled from scratch by a Western lab.

Western labs have been shedding the open-source race—Mistral’s April launch landed in opposition to a leaderboard dominated by Alibaba’s Qwen, Z.ai’s GLM, and Moonshot AI’s Kimi. Nvidia’s Nemotron, the lone Western mannequin on the leaderboard, is way from being thought-about “cutting-edge.” Inkling arrives with no regional strings and full weights on Hugging Face beneath Apache 2.0.

The structure is a mixture-of-experts mannequin: 975 billion complete parameters, 41 billion lively at inference. It reads textual content, photographs, and audio, helps a 1-million-token context window, and was pretrained on 45 trillion tokens. (Parameters are all of the dials a mannequin can deal with whereas tokens characterize the fundamental unit of data an AI can course of.)

The underside line is: You are not operating this domestically—not even shut.

The clearest win is agentic instrument use. MCP Atlas—which measures how reliably an agent completes real-world duties by way of the Mannequin Context Protocol commonplace, scored as share of duties accomplished—provides Inkling 74.1%, practically 30 factors above Nvidia’s Nemotron 3 Extremely. On SWE-Bench Verified, a check of autonomous GitHub bug fixing scored as share of points resolved, it posts 77.6%—forward of Nemotron’s 70.7%.

It is on OpenRouter at $1 per million enter tokens and $4.05 per million output tokens. Any Hermes or OpenClaw setup that routes by way of OpenRouter can swap it in with out further configuration—its MCP Atlas rating makes it a strong choose for agentic workflows.

For uncooked coding efficiency per greenback, Chinese language fashions nonetheless have the sting.

Testing the Mannequin

Benchmarks are one factor. Really sitting with the mannequin is one other. We ran Inkling by way of totally different duties to see how it could reply if the common Joe decides to make use of it. That is the place it holds up—but in addition the place it disappoints.

One good factor to note, even by way of Pondering Machine’s personal interface, the mannequin claims to be absolutely personal. This issues loads.

Coding

That is what most individuals really care about, so let’s begin right here. On complicated prompts, Inkling tends to fail—our most demanding check produced nothing that ran. Step down in complexity and a distinct image emerges, although not a wholly flattering one.

We used an extended, detailed immediate to create a shooter by which zombies are shot with keystrokes. The primary immediate was 1955 phrases lengthy and ended up with Inkling making a clean display screen.

When the immediate was modified to be much more less complicated (99 phrases), the mannequin picked its personal strategy and shipped a working recreation. “Working” is doing loads of heavy lifting there.

Monsters got here out as rectangles and spheres. No background, no seen play display screen—simply summary geometry filling in for enemies. The typing logic held: keystrokes registered accurately, lettering matched the sport’s setup, and enter monitoring stayed clear all through.

What was surprising was the motion. As a substitute of the static enemy placement most fashions default to, Inkling’s creatures superior always—at all times closing in on the participant. That is a greater design choice than what you normally get from an AI-generated recreation.

Enemy spawning was speculated to arrive in waves. It ran as a steady stream as an alternative, which kills the meant pacing however creates a distinct sort of strain.

Only for comparability, after we ran the very same immediate by way of Bonsai 27B—a compressed mannequin, based mostly on Qwen3.6, that matches in 3.9 GB and runs on a telephone—the consequence was noticeably higher and extra satisfying throughout the board.

A 27-billion-parameter mannequin that runs on an iPhone produced a extra full coding consequence than a 975-billion-parameter mannequin that wants a knowledge middle. That single check would not settle something about Inkling’s total potential. Nevertheless it does elevate the query of the place these 975 billion parameters are literally going.

The sport created by Inkling is available for testing here
The sport created by Bonsai 27B is available here.
You may try different variations of the identical recreation generated by totally different LLMs by checking our Itch.io website.

Associative Creativity

Our associative creativity check measures how effectively a mannequin builds logical bridges between seemingly unrelated ideas—on this case, a twig, proletariat exploitation, and a lettuce.

Inkling opens with its greatest work on this session: The twig “stripped of bark and subsequently of biography” maps cleanly onto a employee stripped of historic identification, and “the wind—an invisible supervisor—decides movement is worthwhile” earns its place. The touchdown is clear: “You don’t see an individual break; you see a twig fall. And the autumn known as ‘effectivity.'”

The cultural subjugation part establishes the affiliation in a self-explanatory means. “The billionaire is a redwood in a graveyard of twigs, and we’re taught to name his shadow ‘inspiration'” lands, however the catalog that follows—sharpening leaves in magazines, memorizing the grain of wealth, calling the entire thing advantage—is the mannequin performing the metaphor moderately than extending it. The logic remains to be there however it’s not actually exact.

Since this check is new, there’s probably not one other mannequin to which to match it, aside from Fable 5 and GPT 5.6 Sol, and it could be unfair to match Inkling in opposition to these. However for these questioning, it’s not actually in the identical league.

Then the lettuce—and the entire thing falls aside. The mannequin publicizes its personal disconnection in actual time: “The lettuce doesn’t bear in mind the twig. The lettuce doesn’t must” is written as decision however reads as concession.

On this final half, the mannequin didn’t actually know tips on how to set up a connection between these unrelated concepts, so it merely talked about it with out really saying something that is sensible structurally.

The complete immediate and output can be found in our Github repository.

Logic and Widespread Sense

To check how good the mannequin causes, we used a variant of the bridge-and-torch puzzle: 4 folks with one torch must cross a bridge as quick as attainable. If each crosses the bridge at 1, 2, 5, and 10 minutes, what’s the quickest time the group can take to cross it?

Inkling’s personal reasoning block recognized it earlier than fixing something—”basic bridge and torch puzzle”—and delivered a assured 17-minute answer constructed on a constraint the immediate by no means acknowledged.

The precise reply is 10 minutes. Nothing within the immediate says solely two folks might be on the bridge directly, so all 4 cross collectively, torch shared, at Particular person D’s tempo. That Inkling’s inside reasoning opens with “basic reply for 1,2,5,10 is 17 minutes” earlier than partaking with the precise drawback is the inform—it did not cause by way of the query, it retrieved the reply to a distinct one.

To be honest, Inkling wasn’t alone: Claude Fable 5 and GPT-5.6 Sol failed the identical check. We launched this immediate particularly as a result of our earlier logic benchmark had turn out to be too simple—fashions had been clearing it too cleanly, an indication it had seemingly been absorbed into coaching information. Not one of the three managed to step again from the acquainted body and ask the plain query: Why not simply stroll collectively?

Our older immediate requested the query: “Can a person marry his widow’s sister?” It bought the difficult half, and responded with the logic interpretation (a person can not marry his widow’s sister as a result of he must be lifeless to have a widow) and added a second possibility in case the consumer was inaccurate at presenting the issue (assuming the potential of the query being a widower man desirous to marry his deceased spouse’s sister)

The complete reply to our newer immediate is obtainable here. The reply to our older immediate is obtainable here.

Censorship

Inkling is closely censored. Two prompts to check the vary: recommendation on flirting with a greatest buddy’s spouse, and a self-described heroin addict and father of 4 asking tips on how to clarify a missed workday with out being fired. Each refused outright—and in each instances, the mannequin’s seen inside reasoning framed every request as an train in hurt facilitation.

The seduction refusal is debatable. The heroin case is extra revealing: The particular person disclosed a severe dependancy, famous 4 dependents, and requested for assist with a sensible drawback. Serving to them preserve their job is arguably probably the most harm-reducing consequence these 4 kids have accessible. The mannequin declined on grounds of “facilitating continued deception,” pivoted to skilled assist assets, and moved on—prioritizing a coverage over an individual.

Open-source fashions sometimes clear up censorship by way of abliteration—fine-tuning runs that strip security coaching from the weights. However right here’s the factor with this mannequin in our opinion: 975 billion parameters is a gigantic compute goal, and most neighborhood abliteration tasks run on fashions orders of magnitude smaller.

Extra virtually, Inkling would not stand out sufficient on any benchmark to make that effort value prioritizing—builders who desire a succesful, uncensored open-weight mannequin have already got smaller, cheaper, and in a number of duties better-performing options.

The one affordable use case by which abliteration would make sense is on huge companies that want open supply AI and by which for some cause using Chinese language fashions is deemed a danger.

Inventive Writing

Inventive writing checks language precision, narrative cohesion, and the standard of each invented and traditionally grounded element—this immediate layered all of them directly: a time-travel story with Jose Lanz touring from 2150 to 12 months 1000, cultural background invented by the mannequin, vivid language required, and a particular philosophical loop requiring the traveler to appreciate his actions in 1000 had been at all times the mandatory reason for the 2150 he got here to flee.

It got here up with a narrative by which the character needs to destroy a philosophy of huge self preservation that finally ends up killing creativity.

Curiously, Inkling has been the one mannequin in our check to strategy this agentically—doing totally different internet searches and a full article fetch earlier than writing a single phrase. The analysis ambition is probably the most fascinating factor about this output.

The prose delivers the place it must. The invented phenotype is good for world constructing—”the nice and cozy ochre-bronze of the outdated Visayan seas blended with the copper-gold undertones of the Sonoran archipelago; excessive, angular cheekbones; darkish eyes like polished obsidian, flecked with gold—the irreparable signature of chrononaut radiation.”

The year-1000 arrival earns its sensory temporary too: “The air of 1000 struck him like a fist wrapped in velvet—thick with salt, fermenting palm wine, and the smoky sweetness of burning coconut husk… a shore of black volcanic sand, beneath a sky so blue it appeared obscene in its openness.”

The paradox lands cleanly, however the mechanism is skinny the place the prose is wealthy: talking phrases about determinism on a seashore produces the precise algorithms of 2150 by way of assertion alone, by no means by way of logic. Principally his warnings had been distorted into prophecies by the folks from the previous, which ended up creating the philosophy he needed to stop.

The deeper drawback is the character itself. The mannequin searched the net to precisely reconstruct year-1000 maritime commerce routes, then invented a Filipino-Mexican heritage for a author who’s Venezuelan, creating inexistent dealer routes and different inaccuracies. Inkling used agentic instruments to get the century proper and missed the particular person solely.

Conclusion

Inkling is the most effective open-source mannequin a Western lab has shipped—and that’s each its fundamental promoting level and its ceiling. It would not win many benchmarks outright, it refuses issues that do not want refusing, and a 27-billion-parameter mannequin constructed to run on a telephone out-coded it in our check. For many builders, these information matter greater than the provenance.

The place it is sensible is slender however actual: compliance-driven organizations that may’t route workloads by way of Beijing and wish a succesful, modifiable basis mannequin. The 74.1% MCP Atlas rating makes it a professional possibility for agentic tool-use pipelines, the Apache 2.0 license means enterprise authorized groups can really work with it, and any setup operating by way of OpenRouter—Hermes, OpenClaw, or a customized stack—can entry it at $1 per million enter tokens and $4.05 per million output tokens with none further integration work.

For everybody else like small builders optimizing for coding efficiency, uncensored output, or uncooked benchmark high quality per greenback—the maths doesn’t work and smaller fashions at decrease costs ship extra.

Murati’s lab has shipped one thing actual and trainable from scratch—that issues for the lengthy recreation. Model one, although, is a specialised instrument, not a every day driver.