whatmodel.sh · find the best speech-to-text models you can run

Best multilingual speech-to-text models for a 32GB graphics card

A leaderboard of the best multilingual speech-to-text models for a 32GB graphics card (with 32GB of system RAM). The best one that runs on it right now is SeamlessM4T v2 Large, followed by Cohere Transcribe (03-2026) and Whisper large-v3. Models are ranked on transcription tests in many languages. Models that run on it show the download that fits and the command to run it.

36 models · scores combine 7 benchmarks · updated how it works · find a leaderboard · benchmark your machine
task[all tasks][long-form transcription][multilingual transcription]

Scores use reviewed multilingual transcription evaluations only. Compare scores within this task. Models without task measurements are excluded.

Models for 32GB RAM · 32GB graphics card → ~32GB fast · reset
+ more
card[RTX 5090][RTX 5000 Ada][RTX PRO 4500 Blackwell]

Detect your exact chip or graphics card:

curl -G whatmodel.sh/stt -d max=0 --data-urlencode "mac=$(sysctl -n hw.memsize machdep.cpu.brand_string)"Mac

curl -G whatmodel.sh/stt -d max=0 --data-urlencode "ram=$(awk '/MemT/{print $2}' /proc/meminfo)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>/dev/null)"Linux

curl.exe -G whatmodel.sh/stt -d max=0 --data-urlencode "ram=$((gcim Win32_ComputerSystem).TotalPhysicalMemory)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>$null)"Windows

$ terminal

detecting your OS…

Run curl "whatmodel.sh/stt?ram=32&gpu=card&vram=32"to get this list in your terminal (add &format=json for agents).

ram[8][16][24][32][48][64][96][128][192][256]GB
gpu[apple silicon][graphics card][none]
vram[1][2][4][8][12][16][24][32][48][64][80][96][128][160]GB
price[local][$0.001][$0.002][$0.005][any]max per audio min

detecting your OS…

Score vs API price

#modeltask score↓where↓speed↓RAM↓get itsources
#modeltask score↓where↓speed↓RAM↓get itsources
1Scribe v2 ElevenLabs83±5api $0.0037/audio minelevenlabs.io ↗Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3Open ASR LeaderboardCoSE-E: A Benchmark for Code-switched Speech Evaluation in Enterprise Settings (arXiv:2609.35645v1), Table 2mu-bench: A Multilingual Utterance Transcription Benchmark (arXiv:2609.32082v1), Table 5
2SeamlessM4T v2 Large · fp32 on CPU, fp16 on CUDA (seamless_communication m4t_predict) Meta79±8*local~9.2GBhow to run › Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3
4Universal-3.5 Pro AssemblyAI69±10*api $0.0035/audio minwww.assemblyai.com ↗CoSE-E: A Benchmark for Code-switched Speech Evaluation in Enterprise Settings (arXiv:2609.35645v1), Table 2
5Chirp 3 Google67±16*api price unknowncloud.google.com ↗mu-bench: A Multilingual Utterance Transcription Benchmark (arXiv:2609.32082v1), Table 5
6Voxtral Small 24B (2507) Mistral65±7*api $0.0040/audio min~59GBdocs.mistral.ai ↗Open ASR LeaderboardCoSE-E: A Benchmark for Code-switched Speech Evaluation in Enterprise Settings (arXiv:2609.35645v1), Table 2
7Cohere Transcribe (03-2026) · bf16 Cohere64±9*local~5.5GBhow to run › Open ASR Leaderboard
8Whisper large-v3 · fp16 OpenAI61±6*local~4.2GBhow to run › Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3Open ASR Leaderboard
9Speechmatics Enhanced Speechmatics60±9*api $0.0067/audio minwww.speechmatics.com ↗Open ASR Leaderboard
10Omnilingual ASR LLM 7B v2 · bf16 Meta59±9*local~19GBhow to run › Open ASR Leaderboard
11Nova-3 Deepgram59±10*api $0.0052/audio mindevelopers.cloudflare.com ↗CoSE-E: A Benchmark for Code-switched Speech Evaluation in Enterprise Settings (arXiv:2609.35645v1), Table 2mu-bench: A Multilingual Utterance Transcription Benchmark (arXiv:2609.32082v1), Table 5
12Omnilingual ASR LLM 1B v2 · bf16 Meta59±6*local~6GBhow to run › Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3Open ASR Leaderboard
13Voxtral Mini 3B (2507) · bf16 Mistral57±6*local~12GBhow to run › Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3Open ASR Leaderboard
14Omnilingual ASR LLM 7B · bf16 Meta57±9*local~19GBhow to run › Open ASR Leaderboard
15Omnilingual ASR CTC 1B v2 · bf16 Meta51±6*local~2.9GBhow to run › Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3Open ASR Leaderboard
16Whisper large-v3-turbo · fp16 OpenAI51±9*local~2.4GBhow to run › Open ASR Leaderboard
17Cohere Transcribe Arabic (07-2026) · bf16 (transformers >=5.4.0) Cohere50±9*local~4.1GBhow to run › QCRI Arabic + English ASR leaderboard (Hugging Face Space QCRI/AURA-Arabic-English-ASR-leaderboard)
18Parakeet TDT 0.6B v3 · fp32 NVIDIA49±6local~3.5GBhow to run › Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3Open ASR LeaderboardCoSE-E: A Benchmark for Code-switched Speech Evaluation in Enterprise Settings (arXiv:2609.35645v1), Table 2
19Omnilingual ASR CTC 7B v2 · bf16 Meta49±9*local~16GBhow to run › Open ASR Leaderboard
20Omnilingual ASR LLM 3B v2 · bf16 Meta48±9*local~11GBhow to run › Open ASR Leaderboard
21Phi-4-multimodal-instruct · bf16 Microsoft48±6*local~14GBhow to run › Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3Open ASR Leaderboard
22Qwen3-ASR-1.7B · bf16 Alibaba (Qwen)48±5local~6.1GBhow to run › Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3QCRI Arabic + English ASR leaderboard (Hugging Face Space QCRI/AURA-Arabic-English-ASR-leaderboard)Open ASR Leaderboard
23Omnilingual ASR CTC 3B v2 · bf16 Meta47±9*local~7.9GBhow to run › Open ASR Leaderboard
24Omnilingual ASR CTC 7B · bf16 Meta47±9*local~16GBhow to run › Open ASR Leaderboard
25Omnilingual ASR LLM 3B · bf16 Meta46±9*local~11GBhow to run › Open ASR Leaderboard
26Canary-1B-v2 · onnx-fp32 NVIDIA46±6*local~5.2GBhow to run › Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3Open ASR Leaderboard
27Omnilingual ASR LLM 1B · bf16 Meta44±9*local~6GBhow to run › Open ASR Leaderboard
28Voxtral Mini Transcribe V2 Mistral42±6*†api $0.0030/audio mindocs.mistral.ai ↗Voxtral Realtime
29Omnilingual ASR CTC 3B · bf16 Meta41±9*local~7.9GBhow to run › Open ASR Leaderboard
30Qwen3-ASR-0.6B · bf16 Alibaba (Qwen)38±7*local~2.8GBhow to run › QCRI Arabic + English ASR leaderboard (Hugging Face Space QCRI/AURA-Arabic-English-ASR-leaderboard)Open ASR Leaderboard
31Voxtral Mini 4B Realtime (2602) · bf16 Mistral37±4local~11GBhow to run › Voxtral RealtimeBenchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3Open ASR Leaderboard
32GPT-4o Mini Transcribe OpenAI37±6api $0.0030/audio mindevelopers.openai.com ↗Voxtral Realtimemu-bench: A Multilingual Utterance Transcription Benchmark (arXiv:2609.32082v1), Table 5
33Omnilingual ASR LLM 300M v2 · bf16 Meta35±9*local~4.4GBhow to run › Open ASR Leaderboard
34Omnilingual ASR CTC 1B · bf16 Meta33±9*local~2.9GBhow to run › Open ASR Leaderboard
35Omnilingual ASR LLM 300M · bf16 Meta31±9*local~4.4GBhow to run › Open ASR Leaderboard
36Omnilingual ASR CTC 300M v2 · bf16 Meta14±9*local~1.3GBhow to run › Open ASR Leaderboard
37Omnilingual ASR CTC 300M · bf16 Meta12±9*local~1.3GBhow to run › Open ASR Leaderboard

+ 1 more don't fit your machine or budget · show them

37 of 37 catalogue models listed here · sources and licences

* new or thinly tested: ranked on too few benchmark results · † only the maker’s own numbers so far · ~RAM is estimated; full context may need more memory