Benchmark Almanac

Read-aloud Assessment Benchmark

FastConformer CTC 110M (NeMo) 18.3% tops its leaderboard Gemini 3.6 Flash 54.5% leads the models we track
Not in model cards

Speech models transcribe children reading aloud in Kiswahili. The evaluation measures transcription errors and disagreements with human assessors about which items the child misread.

Fab AI compares 22 model configurations, including Whisper and Gemini, on word error rate and mistake error rate. It published evaluation results in August 2026.

Made by
Fab AI
Released
August 2026
Kind
Speech recognition and translation
Level
Everyday
Scored
Checked automatically
Answers
Open-ended
Measures
Capability
Subject
Education
Language
Swahili
In and out
Audio → Text
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
A host
In views
Audio and speech
Strongest model tested
Gemini 3.8 Flash #24 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026
22
Entries on its own leaderboard, 7 models we track

FastConformer CTC 110M (NeMo) tops the leaderboard with 18.3% Gemini 3.6 Flash leads with 54.5% among models released by September 2026

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. 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. The best model we track is Gemini 3.6 Flash, at 54.5%. This view shows only models we track, each at its best.

The newest model with a score here came out in September 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 FastConformer CTC 110M (NeMo) 18.3%
  2. 2 FastConformer CTC 110M (INT8, Desktop) 23.2%
  3. 3 FastConformer CTC 110M (INT8, Android) 28.8%
  4. 4 MAI-Transcribe-2 45.3%
  5. 5 Gemini 3.6 Flash 54.5%
  6. 6 Gemini 3.8 Flash 55%
  7. 7 OmniASR CTC 3B v2 55.9%
  8. 8 Gemini 3.5 Flash-Lite 66.7%
  9. 9 Gemini 3.7 Flash 72.4%
  10. 10 W2V-BERT 2.0 (Swahili) 72.8%
  11. 11 Sauti ASR 74%
  12. 12 OmniASR LLM 3B v2 (beam) 75.6%
  13. 13 Gemma 4 E2B 75.7%
  14. 14 MMS-1B (Swahili) 75.7%
  15. 15 Gemini 3.1 Flash-Lite 78%
  16. 16 Paza Phi-4 Multimodal 88.2%
  17. 17 Whisper Large (greedy) 174.6%
  18. 18 Whisper Large v2 (greedy) 221.4%
  19. 19 Inkling-Small 257.6%
  20. 20 Whisper Large v2 (beam 5) 399.8%
  21. 21 Whisper Large (beam 5) 403.8%
  22. 22 Paza Whisper Large v3 Turbo 1993.5%

Source: Official Read-aloud Assessment Benchmark leaderboard

Source: Official Read-aloud Assessment Benchmark leaderboard

#EntryScoreSettingDateSource
1 FastConformer CTC 110M (NeMo) open weights, 0.1B 18.3% Official leaderboard
2 FastConformer CTC 110M (INT8, Desktop) open weights, 0.1B 23.2% Official leaderboard
3 FastConformer CTC 110M (INT8, Android) open weights, 0.1B 28.8% Official leaderboard
4 MAI-Transcribe-2 closed weights 45.3% Official leaderboard
5 Gemini 3.6 Flash 54.5% Jul 2026 Official leaderboard
6 Gemini 3.8 Flash 55% Sep 2026 Official leaderboard
7 OmniASR CTC 3B v2 open weights, 3.1B 55.9% Official leaderboard
8 Gemini 3.5 Flash-Lite 66.7% Jul 2026 Official leaderboard
9 Gemini 3.7 Flash 72.4% Aug 2026 Official leaderboard
10 W2V-BERT 2.0 (Swahili) open weights, 0.6B 72.8% Official leaderboard
11 Sauti ASR open weights, 0.8B 74% Official leaderboard
12 OmniASR LLM 3B v2 (beam) open weights, 4.4B 75.6% Official leaderboard
13 Gemma 4 E2B 75.7% Mar 2026 Official leaderboard
14 MMS-1B (Swahili) open weights, 1B 75.7% Official leaderboard
15 Gemini 3.1 Flash-Lite 78% Mar 2026 Official leaderboard
16 Paza Phi-4 Multimodal open weights, 5.6B 88.2% Official leaderboard
17 Whisper Large (greedy) open weights, 1.5B 174.6% Official leaderboard
18 Whisper Large v2 (greedy) open weights, 1.5B 221.4% Official leaderboard
19 Inkling-Small 257.6% Jul 2026 Official leaderboard
20 Whisper Large v2 (beam 5) open weights, 1.5B 399.8% Official leaderboard
21 Whisper Large (beam 5) open weights, 1.5B 403.8% Official leaderboard
22 Paza Whisper Large v3 Turbo open weights, 0.8B 1993.5% Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 3.6 Flash 54.5% Jul 2026 Official leaderboard
2 Gemini 3.8 Flash 55% Sep 2026 Official leaderboard
3 Gemini 3.5 Flash-Lite 66.7% Jul 2026 Official leaderboard
4 Gemini 3.7 Flash 72.4% Aug 2026 Official leaderboard
5 Gemma 4 E2B 75.7% Mar 2026 Official leaderboard
6 Gemini 3.1 Flash-Lite 78% Mar 2026 Official leaderboard
7 Inkling-Small 257.6% Jul 2026 Official leaderboard

Rankings by part

We read 14 parts from the benchmark’s own leaderboard in 2 groups: wer and mer. Each ranks models on one part of the test. Gemini 3.8 Flash leads 4 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Passage

7 models we track have a score here.

  1. 1 Gemini 3.8 Flash ↑1 30.6%
  2. 2 Gemini 3.6 Flash ↓1 34.2%
  3. 3 Gemini 3.5 Flash-Lite 37.3%
  4. 4 Gemini 3.7 Flash 48.9%
  5. 5 Gemini 3.1 Flash-Lite ↑1 52.9%
  6. 6 Gemma 4 E2B ↓1 60%
  7. 7 Inkling-Small 175.4%

Number one over time

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

Google DeepMind
Own leaderboard Not tested since Sep 2026

Source: Official Read-aloud Assessment Benchmark leaderboard

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
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Gemini 3.6 Flash Gemini 3.5 Flash-Lite Gemma 4 E2B 0% 20% 40% 60% 80% 100% 2026 Gemini 3.6 Flash Gemini 3.5 Flash-Lite

Source: Official Read-aloud Assessment Benchmark 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

8 of 9 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark

Catalog record 1 warning 1 pass
  • Warning Task access We don’t know whether the tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (22 entries).
  • Pass Models scored 7 models we track, plus 15 other entries on its leaderboard.
  • Pass Newest score The newest is from September 2026, for Gemini 3.8 Flash.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3.8 Flash (#24).
  • Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
  • Pass Importer We read the board on fab-ai.org: 7 rows for models we track and 15 other entries 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