Hayz Labs · https://hayzlabs.com/reviews/top/voice-chat/
reviews · top 20 list · checked 18 Oct
Top 10 AI voice chat models
Models you talk to out loud and that answer in speech, for voice assistants and phone agents.
Nothing on this list is scored yet: no outside source we can turn into a number covers them. The order below follows the public measure named on each row.
Ordered by an independent speech-to-speech leaderboard (named on each row). One row per model line, newest version only. Open the Table to see the number that placed each one and how it was scored.
- 1Gemini 3.1 Flash LiveGoogle
Most liked in blind voice chats, but Google may shut it down from 17 Nov 2026; move to Gemini 3.8 Live (#2)
Best for Low-cost voice chats that sound natural, for short projects or trialsYou give up Preview that is weaker at multi-step tasks; Google may retire it from 17 Nov 2026
Not scored yetFree - 2Gemini 3.8 LiveGoogle
Among the best-liked voice agent models, free to start and the cheapest of the top group. Weak at hard support tasks.
Best for Voice agents and phone bots that need top quality at a low priceYou give up Finishes only about 30% of hard support tasks; free tier data improves Google products
Not scored yetFree - 3Grok Voice 3.0xAI
Best all-round score on Artificial Analysis tests and finishes most tasks; costs more per minute than Gemini
Best for Voice agents that must finish tasks and handle hard questionsYou give up No free tier listed; at $4.80 an hour, long calls cost more than on Gemini
Not scored yetPay as you go - 4GPT-Live 1OpenAI
Top scores on hard voice agent tasks, but blind listeners prefer Gemini 3.8 Live, which also costs less.
Best for Customer service voice agents that must finish multi-step tasksYou give up Costs more than Gemini 3.8 Live, which blind listeners also prefer
Not scored yetPay as you go - 5gpt-realtime-2.1OpenAI
For apps already built on OpenAI's Realtime API; GPT-Live 1 now ranks higher in blind votes
Best for Voice agents on OpenAI that need to reason and call toolsYou give up Ranks 11th of 22 in blind votes, well behind GPT-Live 1 and both Gemini models
Not scored yetPay as you go - 6Qwen3.5 Omni Plus RealtimeAlibaba
Talks and also sees video; behind the leaders in blind votes, mainly for Alibaba Cloud users
Best for Alibaba Cloud apps that mix voice with video inputYou give up Price not in our files; 18th of 22 in blind votes and no full test score
Not scored yetNo public price - 7Qwen Audio 3.0 Realtime FlashAlibaba
Low-cost voice agent with tool calling, for teams on Alibaba Cloud; mid-table in blind votes
Best for Budget voice agents that call tools or search the web, priced in yuanYou give up 19th of 22 in blind votes; capped at 60 requests a minute
Not scored yetPay as you go - 8gpt-realtime-2.1-miniOpenAI
Cheaper OpenAI realtime option for simple voice tasks; well behind the leaders
Best for Simple voice tasks on OpenAI at about a third of the full model's audio costYou give up 21st of 22 in blind votes, near the bottom of this list
Not scored yetPay as you go - 9Nova 2.5 SonicAmazon
Amazon's voice model on Bedrock, for teams already on AWS; too new to score
Best for Voice apps built on Amazon Bedrock that want to stay on AWSYou give up Not scored yet: no independent test, and price not in our files
Not scored yetNo public price - 10StepAudio 3 RealtimeStepFun
Top reasoning score on spoken questions and free while in preview; not yet in blind voice tests
Best for Trying a realtime voice model that answers spoken questions well, free for nowYou give up Preview that will be retired when paid pricing arrives; no blind voice test yet
Not scored yetFree
| # | Name | Score | What placed it | Price | Best for | You give up |
|---|---|---|---|---|---|---|
| 1 | Gemini 3.1 Flash LiveGoogle · Closed | Not scored | Ranked #1 of 22 in a public blind vote for voice chat (Artificial Analysis)NotesTheir figure: AA Speech Agent Arena Elo 1100. AA Speech Agent Arena #1 of 22 distinct (27 rows), range 1-2; AA Speech to Speech Index #8 of distinct, one step, no change Report a mistake | Free tier; API audio $0.005/min in + $0.018/min out (preview) | Low-cost voice chats that sound natural, for short projects or trials | Preview that is weaker at multi-step tasks; Google may retire it from 17 Nov 2026 |
| 2 | Gemini 3.8 LiveGoogle · Closed · our review | Not scored | Ranked #2 of 22 in a public blind vote for voice chat (Artificial Analysis)NotesTheir figure: AA Speech Agent Arena Elo 1088. AA Speech Agent Arena #2 of 22 distinct, range 1-4; AA Speech to Speech Index #3 of distinct (Extended Thinking High 82.6), no change; matches review Report a mistake | Free tier; API audio $0.005/min in + $0.018/min out | Voice agents and phone bots that need top quality at a low price | Finishes only about 30% of hard support tasks; free tier data improves Google products |
| 3 | Grok Voice 3.0xAI · Closed | Not scored | Ranked #3 of 22 in a public blind vote for voice chat (Artificial Analysis)NotesTheir figure: AA Speech Agent Arena Elo 1070; S2S Index 86.9. AA Speech Agent Arena #3 of 22 distinct, range 2-6; AA Speech to Speech Index #1 of distinct (86.9), no change Report a mistake | API $0.08/min ($4.80/hr) | Voice agents that must finish tasks and handle hard questions | No free tier listed; at $4.80 an hour, long calls cost more than on Gemini |
| 4 | GPT-Live 1OpenAI · Closed · our review | Not scored | Ranked #4 of 22 in a public blind vote for voice chat (Artificial Analysis)NotesTheir figure: AA Speech Agent Arena Elo 1057. AA Speech Agent Arena #4 of 22 distinct (best variant Sol low, board #5), range 3-7; AA S2S Index #2 of distinct (Astra 83.2), one step, no change; matches review Report a mistake | API $0.05/min session fee + backend model billed separately | Customer service voice agents that must finish multi-step tasks | Costs more than Gemini 3.8 Live, which blind listeners also prefer |
| 5 | gpt-realtime-2.1OpenAI · Closed | Not scored | Ranked #11 of 22 in a public blind vote for voice chat (Artificial Analysis)NotesTheir figure: AA Speech Agent Arena Elo 941; S2S Index 75.2. AA Speech Agent Arena #11 of 22 distinct (Minimal); AA S2S Index #5 of distinct (High 75.2), two steps up, moved one step Report a mistake | API audio $32 in / $64 out per 1M tokens | Voice agents on OpenAI that need to reason and call tools | Ranks 11th of 22 in blind votes, well behind GPT-Live 1 and both Gemini models |
| 6 | Qwen3.5 Omni Plus RealtimeAlibaba · Closed | Not scored | Ranked #18 of 22 in a public blind vote for voice chat (Artificial Analysis)NotesTheir figure: AA Speech Agent Arena Elo 861. AA Speech Agent Arena #18 of 22 distinct; not on AA S2S Index (no full score), no adjustment Report a mistake | Not confirmed | Alibaba Cloud apps that mix voice with video input | Price not in our files; 18th of 22 in blind votes and no full test score |
| 7 | Qwen Audio 3.0 Realtime FlashAlibaba · Closed | Not scored | Ranked #19 of 22 in a public blind vote for voice chat (Artificial Analysis)NotesTheir figure: AA Speech Agent Arena Elo 852. AA Speech Agent Arena #19 of 22 distinct; AA S2S Index #12 of distinct (67.7), no change; newer 3.1 Plus tier not on the board yet Report a mistake | API audio CNY 6.78 in / 13.64 out per 1M tokens (Singapore) | Budget voice agents that call tools or search the web, priced in yuan | 19th of 22 in blind votes; capped at 60 requests a minute |
| 8 | gpt-realtime-2.1-miniOpenAI · Closed | Not scored | Ranked #21 of 22 in a public blind vote for voice chat (Artificial Analysis)NotesTheir figure: AA Speech Agent Arena Elo 844. AA Speech Agent Arena #21 of 22 distinct; AA S2S Index #14 of distinct (54.4), one step, no change Report a mistake | API audio $10 in / $20 out per 1M tokens | Simple voice tasks on OpenAI at about a third of the full model's audio cost | 21st of 22 in blind votes, near the bottom of this list |
| 9 | Nova 2.5 SonicAmazon · Closed | Not scored | Score is for Nova 2.0 Sonic, in a public blind vote (Artificial Analysis)NotesTheir figure: AA Arena Elo 982 (Nova 2.0 Sonic). Not on AA Speech Agent Arena yet (GA 5 Oct 2026). Report a mistake | Not confirmed | Voice apps built on Amazon Bedrock that want to stay on AWS | Not scored yet: no independent test, and price not in our files |
| 10 | StepAudio 3 RealtimeStepFun · Closed | Not scored | Got every question right in a spoken reasoning test (Artificial Analysis)NotesTheir figure: AA Big Bench Audio 100%. Not on AA Speech Agent Arena and no full AA S2S Index score, so no board score; preview model ID Report a mistake | Free preview (limited time) | Trying a realtime voice model that answers spoken questions well, free for now | Preview that will be retired when paid pricing arrives; no blind voice test yet |
Nothing matches all of that. , or try everything.
How we ranked this
Order. Ordered by an independent speech-to-speech leaderboard (named on each row). One row per model line, newest version only.
How we score this list. No score yet: we have not set a rule for this leaderboard. The order is the leaderboard's. The full rules are on How we score. Open "How it was scored" on any row to see the working.
What we did not do. We did not run any of these ourselves. Every number comes from the public source linked on its row, checked on 18 Oct 2026.
Spot a mistake? A wrong rank, price or link: tell us and we will check it against the source. Corrections are dated in News.
Reviews are drafted with help from AI models, including Anthropic's Claude. Model scores come only from outside leaderboards and independent evaluations, by a written rule.