ElevenLabsDeepgramAudio matchupSpeech output
Eleven Flash v2.5vs.Aura-2
Which model is better?
Aura-2 is slightly ahead in independent tests, but the lead is not certain.
Pick Eleven Flash v2.5 for
Pick Aura-2 for
Shared tests
Independent evaluators and leaderboards only, the newest shared snapshot per evaluator. With 95 % ranges a lead counts when they do not overlap; for hit rates, when it holds in a 95 % test. Without any uncertainty figure at most “slightly ahead”.
Coval
depends on the subtestEleven Flash v2.5 is clearly ahead in 0 of 20 subtests and slightly in 9, Aura-2 in 0 clearly and slightly in 5. 6 are level.
Coval TTFA leading silence maximum
lower is better, as of November 6, 2026
Coval TTFA leading silence mean
lower is better, as of November 6, 2026
Coval TTFA leading silence p50
lower is better, as of November 6, 2026
Show all 20 subtests
| Subtest | Eleven Flash v2.5 | Aura-2 | Verdict |
|---|---|---|---|
| Coval TTFA leading silence maximum (lower is better) | 116 ms | 534 ms | Eleven Flash v2.5 slightly ahead |
| Coval TTFA leading silence mean (lower is better) | 42 ms | 192 ms | Eleven Flash v2.5 slightly ahead |
| Coval TTFA leading silence p50 (lower is better) | 42 ms | 133 ms | Eleven Flash v2.5 slightly ahead |
| Coval Time to First Audio (TTFA) p90 (lower is better) | 248 ms | 496 ms | Eleven Flash v2.5 slightly ahead |
| Coval Time to First Audio (TTFA) p75 (lower is better) | 218 ms | 433 ms | Eleven Flash v2.5 slightly ahead |
| Coval Time to First Audio (TTFA) p95 (lower is better) | 279 ms | 536 ms | Eleven Flash v2.5 slightly ahead |
| Coval Time to First Audio (TTFA) mean (lower is better) | 204 ms | 307 ms | Eleven Flash v2.5 slightly ahead |
| Coval Time to First Audio (TTFA) maximum (lower is better) | 555 ms | 808 ms | Eleven Flash v2.5 slightly ahead |
| Coval TTFA roundtrip component mean (lower is better) | 162 ms | 115 ms | Aura-2 slightly ahead |
| Coval Time to First Audio (TTFA) p50 (lower is better) | 197 ms | 277 ms | Eleven Flash v2.5 slightly ahead |
| Coval TTFA roundtrip component p50 (lower is better) | 151 ms | 129 ms | Aura-2 slightly ahead |
| Coval TTFA roundtrip component maximum (lower is better) | 517 ms | 464 ms | Aura-2 slightly ahead |
| Coval Word Error Rate (WER) per-sample maximum (lower is better) | 75.0 % | 68.0 % | Aura-2 slightly ahead |
| Coval Word Error Rate (WER) per-sample p95 (lower is better) | 16.0 % | 15.2 % | Aura-2 slightly ahead |
| Coval Word Error Rate (WER) per-sample p90 (lower is better) | 9.1 % | 8.8 % | level |
| Coval Time to First Audio (TTFA) p25 (lower is better) | 182 ms | 178 ms | level |
| Coval pooled corpus Word Error Rate (WER) (lower is better) | 1.9 % | 1.9 % | level |
| Coval Word Error Rate (WER) per-sample p25 (lower is better) | 0.0 % | 0.0 % | level |
| Coval Word Error Rate (WER) per-sample p50 (lower is better) | 0.0 % | 0.0 % | level |
| Coval Word Error Rate (WER) per-sample p75 (lower is better) | 2.7 % | 2.7 % | level |
Source · latest November 6, 2026
Cekura
Aura-2 slightly aheadAura-2 is clearly ahead in 0 of 8 subtests and slightly in 4, Eleven Flash v2.5 in 0 clearly and slightly in 2. 2 are level.
Cekura TTS hard-part accuracy: codes and addresses
as of October 19, 2026
Cekura TTS non-streaming latency
lower is better, as of October 19, 2026
Cekura TTS streaming latency median
lower is better, as of October 19, 2026
Show all 8 subtests
| Subtest | Eleven Flash v2.5 | Aura-2 | Verdict |
|---|---|---|---|
| Cekura TTS hard-part accuracy: codes and addresses | 55.0 % | 87.0 % | Aura-2 slightly ahead |
| Cekura TTS non-streaming latency (lower is better) | 182 ms | 281 ms | Eleven Flash v2.5 slightly ahead |
| Cekura TTS streaming latency median (lower is better) | 177 ms | 126 ms | Aura-2 slightly ahead |
| Cekura TTS hard-part accuracy: numbers and shorthand | 55.0 % | 71.0 % | Aura-2 slightly ahead |
| Cekura TTS ASR-based hard-part accuracy estimate | 64.4 % | 77.2 % | Aura-2 slightly ahead |
| Cekura TTS hard-part accuracy: medical terms | 100.0 % | 86.0 % | Eleven Flash v2.5 slightly ahead |
| Cekura UTMOS-predicted naturalness | 4 | 4 | level |
| Cekura TTS hard-part accuracy: read from context | 75.0 % | 75.0 % | level |
Source · latest October 19, 2026
All data side by side
Prices are verbatim from the provider, because audio models bill per minute, character or token.
| Eleven Flash v2.5 | Aura-2 | |
|---|---|---|
| Task | Speech output | Speech output |
| Developer | ElevenLabs | Deepgram |
| Status | Active | Active |
| Weights | proprietary | proprietary |
| Runs locally | no | no |
| Popularity (out of 100) | 22 | 44 |