Scores by kind of claim
Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.
-
Factual claims
Statements about the present or the past, checked against the record: did they hold up?
67 61 73 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
60 44 76 -
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
Insufficient record: 0 promises resolved, 5 needed for a score.
All types combined: 66 (likely 61 to 72, high confidence), from 105 claims: 94 factual claims, 11 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Marc Andreessen's claims
20 claims
“But in fact, it's building its own AI model.”
Marc Andreessen, a16z co-founder 7 January 2026 2026 Early Interview with Marc Andreessen: AI Revolution Just Started (36Kr)
Cursor, though it started by calling large models, is building its own AI model.
Resolved true The record shows it happened or was so.
What would count as true or false
True: by 2026-01-07, Cursor (Anysphere) had publicly announced, released or credibly been reported to be developing its own AI model(s), meaning models it trains or fine-tunes itself (for example, its own coding or agent model), while it had started by relying on third-party large models. Mostly true: it was building its own models, but with a minor difference in detail (for example, smaller specialised models, or fine-tunes of open-weight base models, rather than a frontier model built from scratch, where the claim does not depend on this). Mixed: the evidence is ambiguous, for example only small auxiliary models plus reports of plans. Mostly false: it had only announced intentions or hiring, with no model in development or released, or it had only re-branded third-party models. False: credible evidence shows Cursor relied entirely on third-party models and was not developing its own.
Evidence
- Introducing Cursor 2.0 and Composer · Cursor Cursor (Anysphere) · 29 October 2025
“Today, we’re releasing two big updates that make Cursor the best place to work with agents: our first coding model, Composer, and a new interface for working with many agents in parallel.”
- Why Cursor's CEO believes OpenAI, Anthropic competition won't crush his startup | TechCrunch TechCrunch · 9 December 2025
“Cursor also confirmed the existence of those models in November when it said in a blog post, “Our in-house models now generate more code than almost any other LLMs in the world.””
- Exclusive: The high costs and thin margins threatening AI coding startups TechCrunch · 7 August 2025
“And Anysphere announced in January that it is attempting to build its own model, which could give it more control over its expenses.”
- A technical report on Composer 2 · Cursor Cursor (Anysphere) · 27 March 2026
“The report covers the full training process, from continued pretraining on an open base model, Kimi K2.5, through large-scale reinforcement learning, with a focus on closely emulating the real Cursor environment.”
Reasoning
E3 (TechCrunch, Aug 2025) reports that Anysphere announced in January 2025 that it was attempting to build its own model. In E1 (Oct 2025), Cursor announced the release of Composer as "our first coding model." In E2 (TechCrunch, Dec 2025), Cursor is quoted saying its in-house models generate more code than almost any other LLMs. Together these show Cursor was building and had released its own trained models before 2026-01-07, which meets the true criterion.
Cite this claim
Marc Andreessen, “Cursor, though it started by calling large models, is building its own AI model.” (said 7 January 2026). Verdict: resolved true. Reliability API, claim c_f93174e58bca, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_f93174e58bca
Claims without a verdict
Is a verdict here wrong, or do you speak for Marc Andreessen? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.