Language models for voice compared
In a chained voice agent, a language model decides what to say next. Phone agents mostly use the fast, low-cost tier of each family, because every turn waits on it.
Too few independent figures to compare these side by side yet. List prices are on the second tab.
- Single figures
- 1 of 8Each from a different test, not comparable
- Vendor claims
- 4 of 8Publish their own figure
- Lowest price
- $0.0005Anthropic Claude Haiku 5.5
- Highest price
- $0.0016Google Gemini 3.5 Flash-Lite
8 compared, read 29 October 2026 or later. Click a bar to open its row, a column heading to sort, ↗ for the source. Tick Compare on up to four rows.
Showing 8 of 8
Too few independent figures to compare yet
Too few to compareNo independent source has measured enough of these 8 the same way to chart them side by side. Below: the single figures there are, who measures them elsewhere, and 4 vendor claims, folded and unranked. Price is on the next tab.
Measured by others, one at a time
Each from a different test, so they cannot be compared with each other.
- Llama 3.1 8B on Groq (tested Llama 3.1 8B Instruct on BFCL’s hosting) 25.83%, overall function-calling accuracy, higher is better. Berkeley Function Calling Leaderboard (BFCL v4). gorilla.cs.berkeley.edu
Measured elsewhere
No independent comparison found.
What RingTru will measure
Time to the first word and whether the model calls the right tool, inside a live call.
What each vendor says about its speed (4, unranked)
Each vendor’s own words, in name order. They time different things, so they are not charted or ranked.
- Anthropic Claude Haiku 5.5 Anthropic ranks it "Fastest" of its four current models platform.claude.com
- GPT OSS 120B on Cerebras about 3,000 tokens per second (vendor figure) cerebras.ai
- GPT OSS 120B on Groq 500 tokens per second (vendor figure) console.groq.com
- Llama 3.1 8B on Groq 560 tokens per second (vendor figure) console.groq.com
- Cost a minute, lowestAnthropic Claude Haiku 5.5 $0.0005
- Input price, lowestAnthropic Claude Haiku 5.5 $0.10
- Output price, lowestAnthropic Claude Haiku 5.5 $0.50
Bar length is the figure itself, so the shortest bar is best in every column.
Vendor claims, data regions and compliance are in each row’s Details, in the vendor’s own words. Claims are never charted or ranked. RingTru has measured none of these yet. Add up a whole call in the calculator.
How to read these numbers
- Tokens per minute of call
- A voice agent sends the whole conversation, plus its instructions, to the model on every turn, so input tokens grow as the call goes on. The calculator asks for your own estimate rather than guessing one.
- Speed matters more than size
- A model that answers in 300 milliseconds beats a cleverer one that takes two seconds, for most calls. Few vendors publish response times, so the arena shows prices only.
Go deeper
Your agent, tested on real calls.
- test calls
- 506
- passed
- 96.8%
- median reply
- 0.94 s
Holds the new day when a caller changes their mind.