Aller au contenu

SpeechifyAI Simba 3.2 Speechify · text-to-speech (TTS) API

Fast enough

106 ms median and 172 ms P95 time to first audio on Coval (30 days to October 22, 2026): Fast enough speed tier.

Benchmark summary: Fast enough speed tier on Coval's P95 (fast enough for most agents). On the Coval and Artificial Analysis frontier: no other model is both faster and rated higher.

SpeechifyAI Simba 3.2 is a proprietary text-to-speech (TTS) API from Speechify, measured by three independent benchmarks: Artificial Analysis, Coval and Speko.

Speed tier

Fast enough

P95 from 138 ms to 263 ms. Fast enough for most agents.

Median time to first audioCoval, 30 days
106 ms
P95 time to first audio (slow turns)Coval, 30 days
172 ms
Naturalness EloSpeko
1573
Listener Elo (95% interval)Artificial Analysis
1,245 ±12
Numbers and dates (0 to 1)Speko
0.21
List price per 1M charactersall sources
$6.6 to $10

Data read October 22, 2026. No combined score: methodology

Gradium TTS vs SpeechifyAI Simba 3.2 Fastest TTS APIs, by speed tier

Speed

On Coval's 30-day window ending October 22, 2026 (3,368 runs from US East), SpeechifyAI Simba 3.2 starts speaking after 105.8 ms on a typical turn (median) and 172.4 ms on a slow turn (P95, the slowest one in twenty). That P95 is 7th lowest of the 32 models Coval measures, which puts it in the Fast enough tier: fast enough for most agents. On the median it is a tie (gap under 10 ms) with Alibaba Qwen3 TTS 1.7b, Palabra TTS v1 and SpeechifyAI Simba 3.0. Speko's board plots it at 56 ms time to first audio from US East; its model page reports a 85 ms sweep.

Voice quality

Speko rates its naturalness at 1,573 Elo from blind A/B votes on phone-agent lines (12th highest of the 43 models Speko rates). In the Artificial Analysis Speech Arena it holds 1,245 ±12 Elo (95% interval), 10th highest of the 98 models Artificial Analysis rates; scores whose intervals overlap are a tie. In its Customer Service category, Artificial Analysis places it 15th, at 1,248 Elo.

Numbers and dates

Speko's pronunciation test on numbers, dates and currency scores it 0.21 on a 0-to-1 scale (23rd highest of the 30 models Speko scores). This is the accuracy measure used here; Coval's word error rate separates models too little to compare them.

Price

Coval lists $10 per 1M characters (since Sep 22, 2026). Artificial Analysis lists $6.6. Speko lists about $10. List prices, as each source prints them: a range of $6.6 to $10.

Benchmark measurements

Latency and accuracy — Coval (30-day window)

Mean TTFA
124.3 ms
P50 / median time-to-first-audio (TTFA)
105.8 ms
P90 TTFA
148.6 ms
P95 TTFA
172.4 ms
Latency consistency (standard deviation)
147 ms
Leading silence
12.2 ms
Time to first byte (network arrival)
112 ms
Accuracy: word error rate (WER)
2.3%
WER basis
pooled
Latency runs
3,368 runs
Accuracy runs
3,371 runs
Window
30d
Coval snapshot
2026-10-22T14:00:19.256741Z
Coval model ID
simba-3.2
Coval model page
benchmarks.coval.ai
Feature tags
emotion-control, voice-cloning
List price per 1M characters
10 USD
Price listed since
Sep 22, 2026

Speko

P50 TTFA, US East
85 ms
Naturalness (Elo)
1,573
Head-to-head win rate
63%
Pronunciation robustness (0 to 1)
0.21
Voice drift (lower is steadier)
7
Price per 1M characters
10 USD
Measured on
2026-07-03
Speko model
simba-3.2
Speko model page
benchmarks.speko.ai
Time to first audio, board chart
56 ms

Artificial Analysis Speech Arena

Customer service rank
15
Customer service Elo
1,248
Customer service Elo 95% interval (±)
25
Speech Arena Elo (blind listener preference)
1,245
Elo 95% interval (±)
12
Arena rank
10
Arena rank range
8-10
Arena appearances
3,367
Voices tested
8
Price per 1M characters
6.6 USD
Released
Jul 2026
Artificial Analysis model
Simba 3.2
Artificial Analysis leaderboard
artificialanalysis.ai

Price (all sources)

Lowest list price per 1M characters
6.6 USD
Highest list price per 1M characters
10 USD

Frontiers

On the Coval and Artificial Analysis frontier
✓

Similar TTS models