Inflection AI

weights

closed

data

closed

code

closed

license

Proprietary(API/enterprise-licensed, no weights)

Inflection offers its models only through its API and enterprise licenses, with no downloadable weights. Models fine-tuned for a customer belong to that customer, but the base weights, training data and code stay with Inflection.

• https://inflection.ai/blog/enterprise recorded 2026-09-29

  Vendor launch post, October 7, 2024: Inflection 3.0 is offered as a hosted system and developers reach the model through the "Large Language Model through its new commercial API". Searched the body for download, weights, open source and GitHub: none.

• https://inflection.ai/labs recorded 2026-09-29

  Inflection AI Labs page describes internal work on "fine-tuning pipeline optimizations" and "distillation from our proprietary foundation models". Searched the body for GitHub, repository, open source, download and Hugging Face: none.

• https://inflection.ai/training-data-transparency-statement recorded 2026-09-29

  Vendor training-data statement, last updated December 22, 2025: the company "trains language models using publicly accessible, licensed, and proprietary datasets" and describes them only in summary. No dataset or data-preparation scripts are linked from the page.

• https://newsroom.intel.com/artificial-intelligence/inflection-ai-intel-launch-enterprise-ai-system recorded 2026-08-13

  PRIMARY - "With Inflection for Enterprise, built on Inflection 3.0, enterprise customers can now harness a comprehensive AI solution ... specifically trained on their unique company data", dated October 7, 2024, delivered via Intel Tiber AI Cloud and a Gaudi 3 appliance. No weights, corpus or code released.

• https://www.maginative.com/article/inflection-ai-unveils-enterprise-offering-with-new-inflection-3-0-models-2/ recorded 2026-09-29

  Maginative launch report, October 7, 2024: "Inflection 3.0, the core of the enterprise offering, comprises two distinct models" sold through a paid API, and "Inflection for Enterprise introduces a proprietary fine-tuning system". Searched the body for Hugging Face, download and weights: none.

After Microsoft's 2024 acqui-hire of most of the research team, Inflection pivoted to enterprise API access, but no usage of any kind -- users, customers or revenue -- has since been disclosed. With no published figure to read from, the reading here is a directional placement rather than one confirmed by a measured figure.

• https://en.wikipedia.org/wiki/Inflection_AI recorded 2026-08-13

  March 2024 - "Microsoft acqui-hiring nearly the entirety of its 70-person workforce"; "As part of the deal, Microsoft paid Inflection $650 million to license its technology". No user, customer or revenue figure published.

Inflection 3.0 was built for emotional intelligence and productivity rather than leaderboard performance, and was never a frontier model even at release. Development halted after Microsoft took most of the research team, and no independent benchmark has appeared since, leaving it far behind the 2026 frontier models.

• https://openrouter.ai/inflection/inflection-3-productivity recorded 2026-09-29

  OpenRouter listing for Inflection 3 Productivity: "Inflection 3 Productivity was released on October 11, 2024" with an 8,000-token context window; no newer Inflection model is listed beside it, only Inflection 3 Pi.

• https://www.maginative.com/article/inflection-ai-unveils-enterprise-offering-with-new-inflection-3-0-models-2/ recorded 2026-09-29

  Maginative describes the two Inflection 3.0 models by use case - the Productivity model "is optimized for following instructions" - and states that "Both models feature an 8k context window". No benchmark result is reported.