Multimodal AI benchmark leaderboard (November 2026)
As of 13 Nov 2026, BenchLeader tracks 23 multimodal AI benchmarks from 9 publishers, 4 of which feed the BenchLeader Index. Across multimodal benchmarks, Gemini 2.5 Deep Think (Google) leads with a multimodal score of 73.0, ahead of Gemini 4 Carbon (71.8) and Gemini 4 Argon (70.9). On the most-weighted live board, MMMU-Pro is led by Gemini 4 Carbon (medium) at 89.3%. Of the 23 boards, 12 are active, 7 near-saturated and 4 retired (retired: no new result in a year).
Last updated · rebuilt every six hours · independent publishers only, no launch-post numbers
- benchmarks
- 23
- active
- 12
- saturated or near
- 7
- publishers
- 9
Other categories: AI agent benchmarks · AI coding benchmarks · Reasoning and math benchmarks · all LLM benchmarks
Which models lead multimodal benchmarks
Ranked on the multimodal category of the BenchLeader Index: every board below that feeds the index, standardised (50 is the average model) and averaged. One row per model, at its best reasoning effort; models need results on at least 1 of these boards.
| # | Model | Provider | Multimodal score | Index | Blended $/M |
|---|---|---|---|---|---|
| 1 | Gemini 2.5 Deep Think (thinking) | 73.0 | – | – | |
| 2 | Gemini 4 Carbon (medium) | 71.8 | 65.8 | $4.00 | |
| 3 | Gemini 4 Argon (medium) | 70.9 | 61.6 | $4.00 | |
| 4 | Claude Opus 5.5 (max) | Anthropic | 70.2 | 68.4 | $8.00 |
| 5 | GPT-5.4 Pro | OpenAI | 69.8 | 62.8 | $67.50 |
| 6 | GPT-6 Astra (high) | OpenAI | 68.9 | 66.3 | $20.00 |
| 7 | Claude Fable 5 (high) | Anthropic | 68.0 | 63.9 | $20.00 |
| 8 | GPT-6.1 Sol (xhigh) | OpenAI | 67.6 | 63.6 | $4.00 |
| 9 | Claude Fable 5.1 (max) | Anthropic | 67.5 | 66.6 | $20.00 |
| 10 | Gemini 3.8 Flash (high) | 67.5 | 61.7 | $1.50 |
Every multimodal benchmark, what it measures and who leads (23)
Multimodal benchmarks pair images, charts, documents, video or audio with questions: college-level visual reasoning (MMMU-Pro), spatial and visual puzzles, video understanding (Video-MME) and arena votes on image prompts.
Use them for: document and chart understanding, screenshots, vision features and audio.
Caveat: The original MMMU can often be answered from the text alone; MMMU-Pro removes those questions and adds ten answer options, so prefer it.
| Benchmark | What it measures | Leader today | Status | Models | Newest result |
|---|---|---|---|---|---|
| MMMU-Pro | College-level multimodal understanding. Run by Artificial Analysis. | Gemini 4 Carbon (medium) 89.3% | Near-saturated | 296 | 7 Nov 2026 |
| Arena Vision | Pairwise votes on image-understanding prompts. | Claude Fable 5 (high) 1326 | ActiveJudge-based | 160 | 9 Nov 2026 |
| VISTA | Vision-language understanding. Scale AI. | Gemini 2.5 Pro 54.6% | Active | 57 | 23 Mar 2026 |
| MMMU (validation) | Official MMMU validation-set leaderboard. | GPT-5.1 85.4% | Near-saturatedContamination risk | 50 | 24 Nov 2025 |
| MortgageTaxnot in index | Reading tax certificate images and extracting the figures. Run by Vals AI. | Claude Opus 5 72.1% | Active | 95 | 1 Sep 2026 |
| MMMU-Pro (Vals)not in index | Vals AI's own run of MMMU-Pro. Run by Vals AI. | Claude Fable 5.1 90.6% | Near-saturated | 92 | 1 Sep 2026 |
| GeoBenchnot in index | Identifying where a photo was taken. | Claude Fable 5 (high) 89.0% | Near-saturated | 44 | 21 Sep 2026 |
| Blueprint-Bench 2not in index | Turning photos of rooms into floor plans. | Gemini 4 Carbon 58.6% | Active | 37 | 10 Nov 2026 |
| LiveBench Designnot in index | LiveBench 2.0 design suite. | Claude Fable 5.2 85.1% | Near-saturated | 34 | 6 Nov 2026 |
| AudioMCnot in index | Multiple-choice questions asked and answered in audio. Scale AI. | Gemini 3.8 Flash (high) 60.4% | Active | 30 | 16 Oct 2026 |
| VPCTnot in index | Visual physics comprehension: predicting what happens next in a scene. | Gemini 3 Pro 91.0% | Near-saturated | 30 | 17 Dec 2025 |
| SAGEnot in index | Grading handwritten maths work from images. Run by Vals AI. | Claude Opus 5.5 91.3% | Near-saturated | 29 | 5 Nov 2026 |
| Code2Video-AAnot in index | Artificial Analysis' run of HeyGen's Code2Video Bench: short product and brand videos built in HTML from… | Claude Opus 5.5 (max) 1536 | Active | 28 | 5 Nov 2026 |
| MMMU-Pro (official)not in index | Official MMMU-Pro leaderboard. Shown for reference. | Gemini 3 Pro 81.0% | Active | 28 | 8 Apr 2026 |
| AA-3Dnot in index | Interactive 3D experiences built from a prompt, graded head to head, so the score is an… | Claude Fable 5.2 (max) 1271 | Active | 25 | 9 Nov 2026 |
| AudioMC (text output)not in index | AudioMC answered in text. Scale AI. | Gemini 3.8 Flash (high) 60.4% | Active | 21 | 16 Oct 2026 |
| VTBnot in index | Scale's vision task benchmark across practical image tasks. Scale AI. | Muse Spark 1.1 44.8% | Active | 21 | 9 Jul 2026 |
| VoiceCodeBenchnot in index | Speech recognition of entities across eight workflow domains. Run by Vals AI. | GPT Live Transcribe 67.7% | Active | 20 | 24 Sep 2026 |
| AudioMC (audio output)not in index | AudioMC answered in speech. Scale AI. | GPT Realtime 2 (xhigh) 48.5% | Active | 15 | 11 May 2026 |
| Video-MMEnot in index | Video understanding across durations, without subtitles. | video-SALMONN 2+ 79.7% | Retired | 42 | 18 Jun 2025 |
| ScienceQAnot in index | Science questions with diagrams. A 2022 benchmark, saturated for current models; kept for history. | Phi 3.5 Vision 91.3% | RetiredContamination risk | 23 | 16 Aug 2024 |
| SpatialViz-Benchnot in index | Spatial visualisation: rotation, folding and animation. | Gemini 2.5 Pro 44.7% | Retired | 8 | 17 Jun 2025 |
| MindCubenot in index | Spatial reasoning from multiple views. | Gemma 3 12B 46.7% | Retired | 5 | 22 May 2025 |
Multimodal AI benchmarks: questions people ask
- What are the best multimodal AI benchmarks?
- The most informative multimodal AI benchmarks today are MMMU-Pro (led by Gemini 4 Carbon (medium) at 89.3%), Arena Vision (led by Claude Fable 5 (high) at 1326) and VISTA (led by Gemini 2.5 Pro at 54.6%): each feeds the BenchLeader Index, covers many models and has not been saturated. Use them for: document and chart understanding, screenshots, vision features and audio.
- Which AI model is best on multimodal AI benchmarks?
- As of 13 Nov 2026: Across multimodal benchmarks, Gemini 2.5 Deep Think (Google) leads with a multimodal score of 73.0, ahead of Gemini 4 Carbon (71.8) and Gemini 4 Argon (70.9). A category score standardises every board (50 is the average model, 15 points one standard deviation) and averages them, so one board cannot decide it.
- Are multimodal AI benchmarks comparable across boards?
- Within a board, yes. Across boards, less so: image, chart, document and video tests use different inputs and scoring, and some older sets (the original MMMU) can be partly answered from the text alone. MMMU-Pro removes those questions, so prefer it when two boards disagree.
- Which multimodal AI benchmarks are saturated?
- As of 13 Nov 2026, MMMU-Pro (near-saturated), MMMU (validation) (near-saturated), MMMU-Pro (near-saturated), GeoBench (near-saturated), LiveBench Design (near-saturated), VPCT (near-saturated) and SAGE (near-saturated). A board is saturated when its fifth-best model scores 90% or more and near-saturated when the leader reaches 90% or the fifth-best 80%; both are decided from the data and change as scores do.
- How often is this leaderboard updated?
- BenchLeader re-reads every publisher's results every six hours (runs start at 04:03, 10:03, 16:03 and 22:03 UTC), and this page is rebuilt with each update. Each board's newest result date is in the table; publishers update on their own schedules.
Related
Cite as: BenchLeader, “Multimodal AI benchmark leaderboard (November 2026)”, data as of 13 Nov 2026, https://www.benchleader.com/benchmarks/multimodal.