DeepMind Lyra Voice

NEW HOT

DeepMind · Other

API

DeepMind Lyra Voice is a groundbreaking model specializing in highly accurate scientific speech transcription and complex analytical reasoning from audio data.

TextAudioReasoningMath

Scores 98% on MMLU, competitive across 4 standard benchmarks.

Data aggregated and editorially reviewed by TrendMing.

Benchmark Performance

MMLU98%

HumanEval91%

MATH93%

GPQA95%

Scores sourced from published evaluations. Higher = better.

Quick Specs

Context Window1.2M tokens

Parameters6T

Input pricing$0.007 / 1K audio-tokens

Output pricing$0.02 / 1K audio-tokens

Released2026-10-02

Back to AI Models