Invideo announced that its Agent One placed first in the latest independent benchmark of AI video agents published by Physion Labs. The tweet posted on July 26th, 2026, says the company was not originally included in the study, requested insertion, and now leads the results. (Invideo on X)
Physion Labs released the benchmark in mid‑July 2026, evaluating sixteen quantitative and qualitative metrics across a range of AI video generation tools. According to Invideo’s thread, Agent One ranked #1 on twelve of those metrics and never fell below #2 on any. The company highlights two metric families: an objective score measuring factual correctness (82.3) and a subjective score capturing human preference (62.5). Both topped the leaderboard. The thread links to the updated benchmark table, though the full blog post has not yet been refreshed.
Invideo’s public statement frames the inclusion as a response to an oversight. The tweet notes, “Last week Physion Labs published an independent benchmark of AI video agents. Invideo wasn't in it. We requested them to include us.” The request appears to have been honored, and the company uses the outcome to argue that “demos are marketing. Evals are evidence.”
Agent One is Invideo’s flagship AI video agent, marketed as a tool that can generate coherent narratives over long durations, a challenge highlighted in the thread. In the second tweet, Invideo points to “narrative coherence, cinematic language and production quality” as the hard problems where most agents falter. By positioning itself as strong on these dimensions, Invideo aims to differentiate from competitors that can only produce short, high‑quality clips.
The significance of the benchmark extends beyond a single leaderboard. Independent evaluations—especially ones that blend objective correctness with human‑rated preference—are scarce in the AI video generation space. Most vendors rely on internal demos or selective case studies. By securing a #1 spot on both objective and subjective families, Invideo signals that its system can satisfy both technical accuracy and audience appeal, a combination that investors and enterprise customers often demand.
The competitive landscape includes well‑known players such as Runway, Synthesia, and Meta’s Make‑A‑Video, each offering varying degrees of control over script, visual style, and output length. None of these rivals are mentioned in the thread, but the broader market has seen a surge in funding for AI‑powered video creation tools over the past two years. Analysts have warned that scaling from short clips to full‑length narratives remains an unsolved problem. Invideo’s claim of sustained top‑two performance across all metrics suggests it may have made progress where others have stalled.
Invideo concludes the thread by thanking Physion Labs for the inclusion and noting that an updated blog post is expected. The company frames the benchmark as “table stakes” for the category, implying that future customers will expect such independent validation as a baseline requirement.
While the tweet provides the headline numbers, it does not disclose the full methodology, sample size, or the specific agents tested. Those details will be critical for analysts trying to gauge how transferable the results are to real‑world production pipelines. Until the Physion Labs blog is updated, the community will have to rely on the snapshot shared in the X thread. The announcement underscores a broader trend: AI video startups are increasingly seeking third‑party validation to move beyond hype and prove utility at scale. If Invideo’s Agent One can consistently deliver narrative‑level video content, it may compel larger platforms to integrate its technology or consider acquisition, a pattern observed in previous AI media deals.
Investors will likely watch closely for any follow‑up data releases from Physion Labs, as the benchmark could become a reference point for future funding rounds in the sector. For now, Invideo positions itself as the new leader in a category where objective quality and human preference have often diverged.