Thinking Machines Lab

Thinking Machines Lab is an AI research and product company. It sells Tinker, a cloud API for fine-tuning and training open models, and publishes Inkling and Inkling-Small, open-weights mixture-of-experts models that take text, image and audio input and can be fine-tuned on Tinker.

Supplied by: Crusoe

Press release · 13 Oct 2026

Partner: Crusoe

Press release · 13 Oct 2026

Partner of NVIDIA

Company disclosure · 10 Mar 2026
StageLate-stage
Founded2025
HeadquartersSan Francisco, USA
StatusPrivate
thinkingmachines.ai

Products

Selected offerings — the AI products recorded here, not the vendor's full catalogue.

Inside Tinker

Closest alternatives 5

Alphabet (Google)PublicSacra's competitive analysis lists Google Vertex AI as an enterprise alternative to Tinker's hosted training/serving API.2 of 2 capabilitiesShared product typeLarger scaleSourced competitorServe a model in productionCompareHugging FaceLate-stageSacra's analyst report names Hugging Face's fine-tuning ecosystem in the same competitive paragraph whose closing sentence states these players 'compete directly with Thinking Machines' customization layer'.2 of 2 capabilitiesShared product typeSmaller scaleSourced competitorTrain or fine-tune a modelCompareNVIDIAPublicThinking Machines Lab's own news post (2026-03-10, read 2026-10-22) announces a multi-year strategic partnership with NVIDIA to deploy at least one gigawatt of Vera Rubin systems; it also says NVIDIA has made a significant investment (amount undisclosed).2 of 2 capabilitiesDifferent layerLarger scaleSourced competitorSourced partnerTrain or fine-tune a modelCompareAmazonPublicSacra's analyst report on Thinking Machines states AWS Bedrock, Microsoft Azure AI and Google Vertex AI 'compete directly with Thinking Machines' customization layer' — an explicit direct-competition claim.2 of 2 capabilitiesShared product typeLarger scaleSourced competitorServe a model in productionCompareMicrosoftPublicSacra's analyst report on Thinking Machines names Microsoft Azure AI among the model-agnostic platforms and states they 'compete directly with Thinking Machines' customization layer'.2 of 2 capabilitiesShared product typeLarger scaleSourced competitorServe a model in productionCompare

Similar products 66

How this list is ranked ↗

BaiduPublicTinker and PaddlePaddle both list model training; larger scale (public).2 of 2 capabilitiesShared product typeLarger scaleCompareTogether AILate-stageOverlaps on model inference and model training; similar scale (private).2 of 2 capabilitiesShared product typeCompareAlibabaPublicTinker and Alibaba Cloud Model Studio both list model training; larger scale (public).2 of 2 capabilitiesShared product typeLarger scaleCompareTencentPublicTinker and Tencent Cloud TI Platform both list model training; larger scale (public).2 of 2 capabilitiesShared product typeLarger scaleCompareNot DiamondUndisclosedTinker and Not Diamond both list model training; Not Diamond's scale not recorded.2 of 2 capabilitiesShared product typeCompareEleutherAINonprofitTinker and GPT-NeoX both list model training; EleutherAI is a nonprofit, with no independent scale; ships developer tool rather than the same layer.1 of 2 capabilitiesDifferent layerCompareFireworks AILate-stageTinker and FireOptimizer both list model training; same layer and the same scale band (late-stage private).2 of 2 capabilitiesShared product typeSame scale bandCompareKuaishouPublicTinker and Vanchin both list model training; similar scale (public); ships application, developer tool and 2 more rather than the same layer.2 of 2 capabilitiesDifferent layerCompareMistral AILate-stageTinker and Mistral Forge both list model training; same layer and the same scale band (late-stage private).1 of 2 capabilitiesShared product typeSame scale bandCompareSiliconFlowUndisclosedTinker and SiliconFlow both list model training; SiliconFlow's scale not recorded; ships model API rather than the same layer.2 of 2 capabilitiesDifferent layerCompareTensorWaveUndisclosedTinker and TensorWave Enterprise Suite both list model training; TensorWave's scale not recorded; ships infrastructure service and platform rather than the same layer.2 of 2 capabilitiesDifferent layerCompareBasetenLate-stageOverlaps on model inference and model training; same layer and the same scale band (late-stage private).2 of 2 capabilitiesShared product typeSame scale bandCompareCoreWeavePublicOverlaps on model inference and model training; larger scale (public).2 of 2 capabilitiesShared product typeLarger scaleCompareHumainUndisclosedTinker and HUMAIN COMPUTE both list model training; Humain's scale not recorded.2 of 2 capabilitiesShared product typeCompareSenseTimePublicTinker and SenseCore both list model training; similar scale (public); ships application, hardware and 4 more rather than the same layer.2 of 2 capabilitiesDifferent layerCompareCactus ComputeUndisclosedTinker and Cactus Platform both list model training; Cactus Compute's scale not recorded; ships infrastructure service and platform rather than the same layer.2 of 2 capabilitiesDifferent layerCompareHyperbolicUndisclosedTinker and Hyperbolic On-Demand GPU Compute both list model training; Hyperbolic's scale not recorded; ships infrastructure service and model API rather than the same layer.2 of 2 capabilitiesDifferent layerCompareHyperstackUndisclosedTinker and Hyperstack AI Studio both list model training; Hyperstack's scale not recorded; ships infrastructure service and platform rather than the same layer.2 of 2 capabilitiesDifferent layerCompareBlack Forest LabsLate-stageTinker and Open Weights Licensing both list model training; smaller scale (private); ships application, infrastructure service and 1 more rather than the same layer.1 of 2 capabilitiesDifferent layerSmaller scaleCompareHuaweiUndisclosedTinker and MindSpore both list model training; Huawei's scale not comparable: only a revenue estimate, no market cap or valuation; ships application, developer tool and 1 more rather than the same layer.2 of 2 capabilitiesDifferent layerCompareVolcano EngineUndisclosedTinker and Machine Learning Platform (机器学习平台) both list model training; Volcano Engine's scale not recorded.2 of 2 capabilitiesShared product typeCompareNAVER CloudUndisclosedTinker and AiTEMS both list model training; NAVER Cloud's scale not recorded.2 of 2 capabilitiesShared product typeCompareUnslothUndisclosedTinker and Unsloth both list model training; Unsloth's scale not recorded; ships developer tool rather than the same layer.2 of 2 capabilitiesDifferent layerCompareEdge ImpulseAcquiredTinker and Edge Impulse both list model training; Edge Impulse is acquired, with no independent scale; ships platform rather than the same layer.2 of 2 capabilitiesDifferent layerCompareKakao EnterpriseUndisclosedTinker and Kubeflow both list model training; Kakao Enterprise's scale not recorded; ships developer tool and platform rather than the same layer.2 of 2 capabilitiesDifferent layerCompareBelvedirUndisclosedTinker and Belvedir both list model training; Belvedir's scale not recorded; ships platform rather than the same layer.2 of 2 capabilitiesDifferent layerCompareNebius GroupPublicTinker and Nebius AI Cloud both list model training; larger scale (public); ships developer tool, infrastructure service and 1 more rather than the same layer.2 of 2 capabilitiesDifferent layerLarger scaleCompareVerdaLate-stageTinker and Verda Cloud both list model training; Verda's scale not recorded.2 of 2 capabilitiesShared product typeCompareAMDPublicOverlaps on model inference and model training; larger scale (public); ships application, hardware and 2 more rather than the same layer.2 of 2 capabilitiesDifferent layerLarger scaleCompareNscaleUndisclosedTinker and Nscale AI Cloud both list model training; Nscale's scale not recorded; ships infrastructure service rather than the same layer.2 of 2 capabilitiesDifferent layerCompareNodus ComputeUndisclosedTinker and Nodus both list model training; Nodus Compute's scale not recorded; ships infrastructure service rather than the same layer.2 of 2 capabilitiesDifferent layerComparePrime IntellectLate-stageTinker and Model Factory both list model training; smaller scale (private).2 of 2 capabilitiesShared product typeSmaller scaleCompareModalLate-stageTinker and Modal Notebooks both list model training; ships developer tool and infrastructure service rather than the same layer.2 of 2 capabilitiesDifferent layerSame scale bandCompareSuper Micro ComputerPublicTinker and SuperCloud Developer Console (SDX) both list model training; Super Micro Computer's scale not comparable: only a revenue estimate, no market cap or valuation; ships hardware and infrastructure service rather than the same layer.2 of 2 capabilitiesDifferent layerCompareLambdaLate-stageTinker and Lambda Cloud both list model training; Lambda's scale not recorded; ships infrastructure service and model API rather than the same layer.2 of 2 capabilitiesDifferent layerCompareSimplismartUndisclosedTinker and Simplismart both list model training; Simplismart's scale not recorded; ships platform rather than the same layer.2 of 2 capabilitiesDifferent layerCompareAethirUndisclosedTinker and Aethir Earth both list model training; Aethir's scale not recorded; ships agent platform and infrastructure service rather than the same layer.2 of 2 capabilitiesDifferent layerCompareOraclePublicTinker and Generative AI Capabilities both list model training; larger scale (public).2 of 2 capabilitiesShared product typeLarger scaleCompareRiftenUndisclosedTinker and Riften both list model training; Riften's scale not recorded; ships platform rather than the same layer.2 of 2 capabilitiesDifferent layerCompareOrchestraUndisclosedTinker and Orchestra both list model training; Orchestra's scale not recorded; ships platform rather than the same layer.2 of 2 capabilitiesDifferent layerCompare4ParadigmPublicTinker and 4Paradigm Sage AIOS both list model training; smaller scale (public); ships application, hardware and 1 more rather than the same layer.2 of 2 capabilitiesDifferent layerSmaller scaleCompareDatabricksLate-stageTinker and Databricks Feature Store both list model training; larger scale (private); ships AI agent, agent platform and 6 more rather than the same layer.2 of 2 capabilitiesDifferent layerLarger scaleCompareTuringLate-stageTinker and RL environments both list model training; smaller scale (private); ships agent platform and data service rather than the same layer.1 of 2 capabilitiesDifferent layerSmaller scaleCompareCueBenchUndisclosedTinker and CueBench environments both list model training; CueBench's scale not recorded; ships data service rather than the same layer.1 of 2 capabilitiesDifferent layerCompareEdotEnvUndisclosedTinker and EdotEnv environments both list model training; EdotEnv's scale not recorded; ships data service rather than the same layer.1 of 2 capabilitiesDifferent layerCompareKimptonUndisclosedTinker and Kimpton Koliseum both list model training; Kimpton's scale not recorded.1 of 2 capabilitiesShared product typeCompareIBMPublicTinker and IBM Machine Learning for z/OS both list model training; larger scale (public).2 of 2 capabilitiesShared product typeLarger scaleCompareORO AIUndisclosedTinker and ORO both list model training; ORO AI's scale not recorded; ships data service rather than the same layer.1 of 2 capabilitiesDifferent layerCompareTokenlessUndisclosedTinker and Tokenless EnvCheck both list model training; Tokenless's scale not recorded; ships developer tool rather than the same layer.1 of 2 capabilitiesDifferent layerCompareDatoricUndisclosedTinker and Datoric datasets both list model training; Datoric's scale not recorded; ships data service rather than the same layer.1 of 2 capabilitiesDifferent layerCompareEnactUndisclosedTinker and Enact both list model training; Enact's scale not recorded; ships data service rather than the same layer.1 of 2 capabilitiesDifferent layerCompareDell TechnologiesPublicTinker and Dell AI Factory with NVIDIA both list model training; larger scale (public); ships application, data service and 3 more rather than the same layer.2 of 2 capabilitiesDifferent layerLarger scaleCompareOlam LabsUndisclosedTinker and Olam Labs Evaluations both list model training; Olam Labs's scale not recorded; ships data service rather than the same layer.1 of 2 capabilitiesDifferent layerCompareOoak DataUndisclosedTinker and Ooak Data RL Environments both list model training; Ooak Data's scale not recorded; ships data service rather than the same layer.1 of 2 capabilitiesDifferent layerCompareTwolabsUndisclosedTinker and Twolabs Environments and Datasets both list model training; Twolabs's scale not recorded; ships data service rather than the same layer.1 of 2 capabilitiesDifferent layerCompareSF ComputeUndisclosedTinker and Autoresearch both list model training; SF Compute's scale not recorded; ships platform rather than the same layer.1 of 2 capabilitiesDifferent layerComparePhysicsXLate-stageTinker and PhysicsX Platform both list model training; smaller scale (private); ships platform rather than the same layer.1 of 2 capabilitiesDifferent layerSmaller scaleCompareStableBrowseUndisclosedTinker and StableBrowse Physical AI Data both list model training; StableBrowse's scale not recorded; ships data service rather than the same layer.1 of 2 capabilitiesDifferent layerCompareLuminary CloudLate-stageTinker and Luminary Platform both list model training; Luminary Cloud's scale not recorded; ships platform rather than the same layer.2 of 2 capabilitiesDifferent layerCompareAleph AlphaLate-stageTinker and PhariaStudio both list model training; Aleph Alpha's scale not recorded; ships application, developer tool and 3 more rather than the same layer.2 of 2 capabilitiesDifferent layerCompareUiPathPublicTinker and UiPath AI Center both list model training; similar scale (public).2 of 2 capabilitiesShared product typeCompareKiln AIUndisclosedTinker and Kiln both list model training; Kiln AI's scale not recorded; ships AI agent and developer tool rather than the same layer.1 of 2 capabilitiesDifferent layerCompareYandexPublicTinker and Yandex DataSphere both list model training; similar scale (public).2 of 2 capabilitiesShared product typeCompareLG CorpPublicTinker and EXAONE Data Foundry both list model training; similar scale (public); ships AI agent, application and 1 more rather than the same layer.2 of 2 capabilitiesDifferent layerCompareAshrUndisclosedTinker and Ashr both list model training; Ashr's scale not recorded; ships platform rather than the same layer.1 of 2 capabilitiesDifferent layerCompareTolokaUndisclosedTinker and Toloka Train both list model training; Toloka's scale not recorded; ships AI agent, data service and 1 more rather than the same layer.1 of 2 capabilitiesDifferent layerCompare

Closest alternatives are pairs a vendor page itself compares. Similar products are matched from what each product does; nothing in that group is asserted as a rivalry.

Sources

Total funding

News reporting · 15 Jul 2025

Mira Murati's AI startup Thinking Machines valued at $12 billion in early-stage funding, Reuters, July 15, 2025. Read via stockanalysis.com's citation for this figure on 2026-08-27.

Stage, from valuation

News reporting · 15 Jul 2025

Mira Murati's Thinking Machines Lab is worth $12B in seed round, TechCrunch, 2025-07-15. post-money valuation of the round dated 2025-07-15. Read via stockanalysis.com's citation for this figure on 2026-08-27.

Founded

Wikipedia · 11 Sep 2026

"The company was founded in February 2025."

Description

Official documentation · 27 Oct 2026

Launch post dated Jul 15, 2026: "Our model, called Inkling, is a Mixture-of-Experts transformer with 975B total parameters, 41B active"; "reasons natively over text, images, and audio". Jul 31, 2026 post (thinkingmachines.ai/blog/a-safe-path-to-open-weights/): "we recently released Inkling and Inkling-Small, two open-weight language models". Tinker wording from tinker-docs.thinkingmachines.ai: "A cloud training API that gives you full control over your data, algorithms, and models."

Open Thinking Machines Lab in the directory

Something wrong here? Send a correction — quote this company id: thinking-machines-lab.