GAIK Ajokortti transcription test
Models transcribe one three-minute Finnish dental webinar excerpt with two speakers. The project reports word error rate and separately counts errors that change the meaning.
GAIK compares 26 configurations, including Gemini, Chirp, Whisper, Scribe and Qwen. Correcting 14 disputed reference passages changed the leading word error rates.
- Released
- January 2026
- Kind
- Speech recognition and translation
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Subject
- Clinical medicine
- Language
- Finnish
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Audio and speech, Biology and medicine
- Strongest model tested
-
Gemini 3.1 Pro #40 on the capability index
- Last model added
- July 2026
- Scores refreshed
- November 20th, 2026
Gemini 3.5 Flash leads with 1.00% among models released by July 2026 Gemini 3.5 Flash leads with 1.00% among models released by July 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. This view shows only models we track, each at its best.
The newest model with a score here came out in July 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1
Gemini 3.5 Flash 1.00%
- 2 Google Chirp 3 2.45%
- 3 Whisper large-v3 chunked decoding 3.60%
- 4 ElevenLabs Scribe v2 4.10%
- 5 This project, la-murre version chunked decoding 4.60%
- 6 Whisper large-v3-turbo 5.00%
- 7 This project's pilot: Whisper large-v3 + LoRA on FLEURS and VoxPopuli 5.20%
- 8 GPT-Transcribe 6.35%
- 9 WhisperX + Finnish-NLP whisper-large-finnish-v3 (Haaga-Helia WhisperX server) 6.60%
- 10 OpenAI whisper-1 6.70%
- 11 This project, round 1 chunked decoding 6.70%
- 12 gpt-4o-transcribe 6.75%
- 13 Finnish-NLP whisper-large-finnish-v3 (plain transformers) 6.85%
- 14 This project, round 1: Whisper large-v3 + LoRA on SKN dialects, FLEURS and VoxPopuli 6.95%
- 15 gpt-4o-mini-transcribe 7.30%
- 16 Orukeet (Parakeet v3 fine-tune) 8.45%
- 17 This project, first model (v1.0) chunked decoding 8.80%
- 18 RASMUS Finnish-ASR-Canary-v2, greedy 10.1%
- 19 RASMUS Finnish-ASR-Canary-v2 + KenLM 11.9%
- 20
Gemini 3.1 Pro 12.7%
- 21 NVIDIA Canary 1B v2 13.4%
- 22 NVIDIA Parakeet TDT 0.6B v3 15.1%
- 23 Finnish-NLP wav2vec2 XLS-R 1B (no LM) 16.1%
- 24
Qwen3-ASR-1.7B 27.1%
- 25
Qwen3-ASR-0.6B 34.5%
Source: Official GAIK Ajokortti transcription test leaderboard
- 1
Gemini 3.5 Flash 1.00%
- 2 Whisper large-v3 chunked decoding 3.60%
- 3 Whisper large-v3-turbo 5.00%
- 4 GPT-Transcribe 6.35%
- 5 gpt-4o-transcribe 6.75%
- 6 gpt-4o-mini-transcribe 7.30%
- 7
Gemini 3.1 Pro 12.7%
- 8
Qwen3-ASR-1.7B 27.1%
- 9
Qwen3-ASR-0.6B 34.5%
Source: Official GAIK Ajokortti transcription test leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 1.00% | May 2026 | Official leaderboard | |
| 2 | Google Chirp 3 closed weights | 2.45% | Official leaderboard | ||
| 3 | Whisper large-v3 | 3.60% | chunked decoding | Nov 2023 | Official leaderboard |
| 4 | ElevenLabs Scribe v2 closed weights | 4.10% | Official leaderboard | ||
| 5 | This project, la-murre version open weights | 4.60% | chunked decoding | Official leaderboard | |
| 6 | Whisper large-v3-turbo | 5.00% | Oct 2024 | Official leaderboard | |
| 7 | This project's pilot: Whisper large-v3 + LoRA on FLEURS and VoxPopuli open weights | 5.20% | Official leaderboard | ||
| 8 | GPT-Transcribe | 6.35% | Jul 2026 | Official leaderboard | |
| 9 | WhisperX + Finnish-NLP whisper-large-finnish-v3 (Haaga-Helia WhisperX server) open weights | 6.60% | Official leaderboard | ||
| 10 | OpenAI whisper-1 closed weights | 6.70% | Official leaderboard | ||
| 11 | This project, round 1 open weights | 6.70% | chunked decoding | Official leaderboard | |
| 12 | gpt-4o-transcribe | 6.75% | Mar 2025 | Official leaderboard | |
| 13 | Finnish-NLP whisper-large-finnish-v3 (plain transformers) open weights | 6.85% | Official leaderboard | ||
| 14 | This project, round 1: Whisper large-v3 + LoRA on SKN dialects, FLEURS and VoxPopuli open weights | 6.95% | Official leaderboard | ||
| 15 | gpt-4o-mini-transcribe | 7.30% | Mar 2025 | Official leaderboard | |
| 16 | Orukeet (Parakeet v3 fine-tune) open weights | 8.45% | Official leaderboard | ||
| 17 | This project, first model (v1.0) open weights | 8.80% | chunked decoding | Official leaderboard | |
| 18 | RASMUS Finnish-ASR-Canary-v2, greedy open weights | 10.1% | Official leaderboard | ||
| 19 | RASMUS Finnish-ASR-Canary-v2 + KenLM open weights | 11.9% | Official leaderboard | ||
| 20 | | 12.7% | Feb 2026 | Official leaderboard | |
| 21 | NVIDIA Canary 1B v2 open weights | 13.4% | Official leaderboard | ||
| 22 | NVIDIA Parakeet TDT 0.6B v3 open weights | 15.1% | Official leaderboard | ||
| 23 | Finnish-NLP wav2vec2 XLS-R 1B (no LM) open weights | 16.1% | Official leaderboard | ||
| 24 | | 27.1% | Jan 2026 | Official leaderboard | |
| 25 | | 34.5% | Jan 2026 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 1.00% | May 2026 | Official leaderboard | |
| 2 | Whisper large-v3 | 3.60% | chunked decoding | Nov 2023 | Official leaderboard |
| 3 | Whisper large-v3-turbo | 5.00% | Oct 2024 | Official leaderboard | |
| 4 | GPT-Transcribe | 6.35% | Jul 2026 | Official leaderboard | |
| 5 | gpt-4o-transcribe | 6.75% | Mar 2025 | Official leaderboard | |
| 6 | gpt-4o-mini-transcribe | 7.30% | Mar 2025 | Official leaderboard | |
| 7 | | 12.7% | Feb 2026 | Official leaderboard | |
| 8 | | 27.1% | Jan 2026 | Official leaderboard | |
| 9 | | 34.5% | Jan 2026 | Official leaderboard |
2 models have held number one on the official leaderboard since its release in January 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official GAIK Ajokortti transcription test leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official GAIK Ajokortti transcription test 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
- 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
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (25 entries).
- Pass Models scored 9 models we track, plus 16 other entries on its leaderboard.
- Pass Newest score The newest is from July 2026, for GPT-Transcribe.
- Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3.1 Pro (#40).
- Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 10 rows for models we track and 16 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.