Benchmark Almanac

S-EMBER

Gemini 3.1 Pro 54.4% leads since July 2026
Not in model cards

Models answer memory questions from first-person video using only frames available before the query time. The benchmark scores multiple-choice accuracy and temporal overlap between predicted evidence intervals and annotated intervals.

The release contains 9,448 question-answer pairs over 388 hours of video. The free-text grounding track can also use a Gemini judge for answer correctness. Frame budgets and query-time clipping affect comparability.

Made by
Meta FAIR and Reality Labs
Released
July 2026
Kind
Long videos
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Video → Text
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
In views
Vision
Strongest model tested
GPT-5.4 #28 on the capability index
Last model added
March 2026
Scores refreshed
November 20th, 2026

In the authors’ words

As wearable devices enable continuous first-person recording, AI assistants must reason across long time horizons to recall past experiences-a capability known as episodic memory. Current benchmarks often rely on offline evaluation with access to entire video files, failing to simulate the streaming reality of wearable intelligence. We introduce S-EMBER (Streaming Egocentric Memory Benchmark for Episodic Retrieval), a large-scale benchmark comprising 3,141 videos totaling 388 hours of organic activity captured via Ray-Ban Meta smart glasses. S-EMBER formalizes grounded streaming episodic retrieval, a paradigm shift from global offline search to causal, active recall triggered by visual events in a continuous stream. We provide 9,448 QA pairs requiring manual visual proof through precise temporal localization and supporting flexible response lengths to simulate natural human-AI interaction. Our extensive benchmarking of frontier models reveals a grounded recall gap: models answer and localize with moderate competence in isolation, yet fall furthest short of human performance when both must hold for the same query, the strongest reaching less than half the human rate. S-EMBER establishes a hardware-authentic foundation for developing grounded, reliable episodic memory in the next generation of wearable AI agents.

Abstract of S-EMBER: A Large-Scale Benchmark for Streaming Egocentric Memory Retrieval (arXiv:2607.02689, CC0)
6
Entries in its paper, 5 models we track

Gemini 3.1 Pro leads with 54.4% among models released by March 2026 Gemini 3.1 Pro leads with 54.4% among models released by March 2026

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The headline is five-way multiple-choice accuracy on the streaming memory questions. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

The newest model with a score here came out in March 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Gemini 3.1 Pro Native video at 1 fps 54.4%
  2. 2 Qwen3-VL-32B-Instruct 128 uniformly sampled frames 42.4%
  3. 3 GPT-5.4 Up to 50 images 41.8%
  4. 4 InternVL3.5-38B 128 uniformly sampled frames 36.4%
  5. 5 LLaVA-OneVision 7B 128 uniformly sampled frames 30.5%
  6. 6 GPT-4o (snapshot unspecified) Text only 20.2%

Source: The S-EMBER paper

#EntryScoreSettingDateSource
1 Gemini 3.1 Pro 54.4% Native video at 1 fps Feb 2026 Official leaderboard
2 Qwen3-VL-32B-Instruct 42.4% 128 uniformly sampled frames Oct 2025 Official leaderboard
3 GPT-5.4 41.8% Up to 50 images Mar 2026 Official leaderboard
4 InternVL3.5-38B 36.4% 128 uniformly sampled frames Aug 2025 Official leaderboard
5 LLaVA-OneVision 7B 30.5% 128 uniformly sampled frames Aug 2024 Official leaderboard
6 GPT-4o (snapshot unspecified) 20.2% Text only Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 3.1 Pro 54.4% Native video at 1 fps Feb 2026 Official leaderboard
2 Qwen3-VL-32B-Instruct 42.4% 128 uniformly sampled frames Oct 2025 Official leaderboard
3 GPT-5.4 41.8% Up to 50 images Mar 2026 Official leaderboard
4 InternVL3.5-38B 36.4% 128 uniformly sampled frames Aug 2025 Official leaderboard
5 LLaVA-OneVision 7B 30.5% 128 uniformly sampled frames Aug 2024 Official leaderboard

Rankings by part

We read 6 parts from the benchmark’s paper, each ranking models on one part of the test, such as Clean accuracy, Overall free-form accuracy and Hallucination rate. Gemini 3.1 Pro leads 5 of them, and InternVL3.5-38B leads the other one. Arrows show how far a model moved from its overall rank among the models we track.

Clean accuracy

5 models we track have a score here.

  1. 1 Gemini 3.1 Pro 42.9%
  2. 2 InternVL3.5-38B ↑2 25.5%
  3. 3 Qwen3-VL-32B-Instruct ↓1 24.3%
  4. 4 GPT-5.4 ↓1 23.8%
  5. 5 LLaVA-OneVision 7B 16.6%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind
Benchmark paper Not tested since Jul 2026

Source: The S-EMBER paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3.1 Pro Qwen3-VL-32B-Instruct InternVL3.5-38B LLaVA-OneVision 7B 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3.1 Pro Qwen3-VL-32B-Instruct

Source: Official S-EMBER leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Audio and video

All 695 in Audio and video
  • Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy. Qwen 3.8 Max 88.7%
  • LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization. Gemini 4 Argon 91.7%
  • MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy. Qwen 3.8 Max 82.4%
  • MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy. Qwen 3.8 Max 90.8%
  • FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better. Phi-4-multimodal 4.00%
  • VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules. Qwen3.5-Omni-Plus 93.1%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 8 months ago, in March 2026, for GPT-5.4. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (6 entries).
  • Pass Models scored 5 models we track, plus 1 other entry on its leaderboard.
  • Warning Newest score The newest came 8 months ago, in March 2026, for GPT-5.4.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.4 (#28).
  • Pass Headroom The best is 54.4%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 5 rows for models we track and 1 other entry at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check