Ascending Labs Research
Benchmark checker Vision and generation

Is Video-MME still meaningful?

Status:Near ceiling Checked against the report, 5 November 2026

Verdict

Barely. Top scores reached 85 in 2025, and no dated 2026 leaderboard exists.

Why near ceiling: The chapter rates Video-MME Near ceiling. Gemini 2.5 Pro scored 85.1 without subtitles in November 2025.[1][1] Qwen3-VL technical report, arXiv 2511.21631 Primary

What “Near ceiling” means

Known problems

  • No dated 2026 leaderboard exists, so the top score dates from November 2025.[1][1] Qwen3-VL technical report, arXiv 2511.21631 Primary
  • VideoMMMU sits at a similar level, with 84.6 for GPT-5 (Nov 2025).[1][1] Qwen3-VL technical report, arXiv 2511.21631 Primary

What to look at instead

MVBench tests 20 temporal tasks that one frame cannot solve. Qwen3-VL scored 76.5 there (Nov 2025).[3][3] MVBench, arXiv 2311.17005 Primary[1][1] Qwen3-VL technical report, arXiv 2511.21631 Primary

Where it appears in the report

Chapter 7, Multimodal, voice and embodied

Sources

Copied from the chapter source lists. Each number links back to the source in its chapter.

  1. [1]
    Qwen3-VL technical report, arXiv 2511.21631. Lab report; Table 2 compares four model families. Primary arxiv.org Chapter 7, source 1
  2. [2]
    Video-MME, arXiv 2405.21075. Primary arxiv.org Chapter 7, source 28
  3. [3]
    MVBench, arXiv 2311.17005. Primary arxiv.org Chapter 7, source 29
  • Deepfake-Eval-2024 Status:Still useful Yes, as a reality check. Open-source detectors lose about half their skill on its real fakes.
  • HallusionBench Status:Still useful Yes. Frontier models still score only 60–67, so it separates them on hallucination.
  • GenEval 2 Status:New / unproven Yes, tentatively. The best model reaches only 35.8%, but outside use is still thin.
  • MMMU-Pro Status:Near ceiling Partly. Scores sit near 85% and differ by source, so compare models within one source only.