Skip to content
Model Menu
Models
Compare
Leaderboard
Benchmarks
⌘K
Top
Chat
Coding
Agents
Writing
Vision
Image
Video
Speech
Embeddings
Decision
Music
Filters
Best for Speech & Voice
All models
2314
Chat & Knowledge
563
Coding
383
Agents & Tool Use
543
Writing & Language
361
Vision & Documents
397
Image Generation & Editing
184
Video Generation
148
Speech & Voice
358
Embeddings & Retrieval
688
Decision models
118
Music & Sound
29
Text to speech
136
Speech to text
151
Realtime
55
Translation
8
Voice changers
4
τ-Voice
Speech Agent Arena
EVA-Bench accuracy
Time to first audio
AA Speech-to-Speech Index
Big Bench Audio
τ³-voice · Airline 1.0.0
All benchmarks
34
Speech & Voice benchmarks
34
×
All
34
Agents
21
Multimodal
12
Other
1
Benchmark
By
Leader
Models
Agents
21
τ-Voice
AA · aggregate
Spoken customer-service agents must use tools to complete policy-constrained airline, retail and telecom tasks in multi-turn voice conversations; the score is task success rate
AA
GPT Live 1
74.5%
17
+3 versions & slices
AA · airline
Spoken customer-service agents must use tools to complete policy-constrained airline, retail and telecom tasks in multi-turn voice conversations; the score is task success rate
AA
GPT Live 1
76.7%
17
AA · retail
Spoken customer-service agents must use tools to complete policy-constrained airline, retail and telecom tasks in multi-turn voice conversations; the score is task success rate
AA
GPT Live 1
72.5%
17
AA · telecom
Spoken customer-service agents must use tools to complete policy-constrained airline, retail and telecom tasks in multi-turn voice conversations; the score is task success rate
AA
Gemini 3.8 Live Extended Thinking
83.9%
17
τ³-voice
Sierra · 1.0.1 · standard · overall
Voice agents converse with a simulated customer and use tools to complete overall service tasks under the named simulator
Sierra
GPT Live 1
81.7%
3
+16 versions & slices
Sierra · 1.0.0 · standard · airline
Voice agents converse with a simulated customer and use tools to complete airline service tasks under the named simulator
Sierra
Gemini 3.1 Flash Live Preview
64%
5
Sierra · 1.0.0 · standard · overall
Voice agents converse with a simulated customer and use tools to complete overall service tasks under the named simulator
Sierra
Gemini 3.1 Flash Live Preview
43.8%
5
Sierra · 1.0.0 · standard · retail
Voice agents converse with a simulated customer and use tools to complete retail service tasks under the named simulator
Sierra
GPT Realtime 2
47.4%
5
Sierra · 1.0.0 · standard · telecom
Voice agents converse with a simulated customer and use tools to complete telecom service tasks under the named simulator
Sierra
GPT Realtime 2
21.9%
5
Sierra · 1.0.1 · custom · banking knowledge
Voice agents converse with a simulated customer and use tools to complete banking knowledge service tasks under the named simulator
Sierra
GPT Live 1
37.1%
3
Sierra · 1.0.1 · standard · airline
Voice agents converse with a simulated customer and use tools to complete airline service tasks under the named simulator
Sierra
GPT Live 1
82%
3
Sierra · 1.0.1 · standard · retail
Voice agents converse with a simulated customer and use tools to complete retail service tasks under the named simulator
Sierra
GPT Live 1
78.9%
3
Sierra · 1.0.1 · standard · telecom
Voice agents converse with a simulated customer and use tools to complete telecom service tasks under the named simulator
Sierra
GPT Live 1
84.2%
3
Sierra · 1.0.1 · custom · airline
Voice agents converse with a simulated customer and use tools to complete airline service tasks under the named simulator
Sierra
Grok Voice 3.0
80%
2
Sierra · 1.0.1 · custom · overall
Voice agents converse with a simulated customer and use tools to complete overall service tasks under the named simulator
Sierra
GPT Live 1
86.2%
2
Sierra · 1.0.1 · custom · retail
Voice agents converse with a simulated customer and use tools to complete retail service tasks under the named simulator
Sierra
GPT Live 1
84.2%
2
Sierra · 1.0.1 · custom · telecom
Voice agents converse with a simulated customer and use tools to complete telecom service tasks under the named simulator
Sierra
GPT Live 1
96.5%
2
Sierra · 1.0.0 · custom · airline
Voice agents converse with a simulated customer and use tools to complete airline service tasks under the named simulator
Sierra
GPT Realtime 2
66%
1
Sierra · 1.0.0 · custom · overall
Voice agents converse with a simulated customer and use tools to complete overall service tasks under the named simulator
Sierra
GPT Realtime 2
51.2%
1
Sierra · 1.0.0 · custom · retail
Voice agents converse with a simulated customer and use tools to complete retail service tasks under the named simulator
Sierra
GPT Realtime 2
50%
1
Sierra · 1.0.0 · custom · telecom
Voice agents converse with a simulated customer and use tools to complete telecom service tasks under the named simulator
Sierra
GPT Realtime 2
37.7%
1
Multimodal
10
AA Speech Agent Arena
overall · Elo
Voice agents respond to spoken requests in the overall pool, scored by pairwise response preference
AA
Gemini 3.1 Flash Live Preview
1,095
19
+3 versions & slices
overall · Task Success
Voice agents respond to spoken requests in the overall pool, scored by task completion
AA
Grok Voice 3.0
94.8%
19
agentic · Elo
Voice agents respond to spoken requests in the agentic pool, scored by pairwise response preference
AA
Gemini 3.1 Flash Live Preview
1,079
19
non-agentic · Elo
Voice agents respond to spoken requests in the non-agentic pool, scored by pairwise response preference
AA
Gemini 3.1 Flash Live Preview
1,151
19
EVA-Bench
EVA-A accuracy pass
Voice agents respond in spoken interactions, scored on response accuracy
ServiceNow
Grok Voice 3.0
73.7%
12
+2 versions & slices
EVA-X experience pass
Voice agents respond in spoken interactions, scored on conversational experience
ServiceNow
Gemini 3.8 Live
88.6%
12
response speed
Voice agents respond in spoken interactions, measured by response delay
ServiceNow
Gemini 3.8 Live Extended Thinking
1.6
6
AudioMultiChallenge
audio output · APR
Scale Audio MultiChallenge · audio output · APR: Speech models follow multiple constraints when producing spoken responses, scored by all-pass rate
Scale
GPT Realtime 2
48.5%
4
Time to first audio
AA
Elapsed seconds from the end of the user audio to the first generated response audio, measured by Artificial Analysis
AA
K
Raon SpeechChat 9B
0
25
AA Speech to Speech Index
v2.0
Speech models complete the component spoken tasks combined in the AA speech-to-speech composite
AA
Grok Voice 3.0
86.9
14
Other
1
AA TTS provider voice
People listen to two voices reading the same text and pick the more natural one
AA
GPT Realtime 2
1,076
1
Legacy · saturated or no longer reported
2
Big Bench Audio
AA v1.2
Speech models answer spoken reasoning and knowledge questions in Big Bench Audio 1.2
AA
StepAudio 3 Realtime Preview
99.7%
28
Full-Duplex-Bench
AA subset v1/v1.5
Speech models handle turn taking and interruptions on AA’s stated Full-Duplex-Bench subset
AA
StepAudio 3 Realtime Preview
98.9%
24
No benchmark matches
Model
τ³-voice · Airline 1.0.0
Languages
Price covers
Price
Released
Gemini 3.1 Flash Live Preview
Google
64
high
en, es, fr +97
Provider-stated input/output audio minute components; text, tools and session/backend charges remain separate.
$0.005
/
$0.018
/ minute
Mar 2026
GPT Realtime 2
OpenAI
58
xhigh
—
—
$32
/
$64
/ 1M audio tokens
May 2026
GPT Realtime 1.5
OpenAI
40
—
—
$32
/
$64
/ 1M audio tokens
Feb 2026
GPT Realtime
OpenAI
36
unstated
—
—
$32
/
$64
/ 1M audio tokens
Aug 2025
Gemini 2.5 Flash Live API
Google
30
—
—
$3
/
$12
/ 1M audio tokens
Show 50 not measured on τ³-voice · Airline 1.0.0
