Mercury 2 vs Gemini Live

Choose Mercury 2 for a low-cost, replaceable text layer behind separate speech services.

Choose Gemini 3.1 Flash Live Preview when the model should hear and speak directly.

We do not declare a latency winner because the available measurements time different systems.

“Gemini Live” is the product category. Google's current developer model is gemini-3.1-flash-live-preview, released March 26, 2026. The preview label is a deployment risk, not a footnote.

Side-by-side architecture

Table 1 Download CSV

Question	Mercury 2	Gemini 3.1 Flash Live Preview
Input and output	Text in, text out	Native audio plus text, image, and video input, audio output
Voice design	Cascaded STT → model → TTS	Audio-to-audio Live API
Context	128K	128K for native audio, lower limits can apply to other live modes
Tool use	Supported	Function calling supported
Structured output	Schema-aligned JSON documented	Not supported on the current model page
Search grounding	Build with an external tool	Supported, separately priced after allowance
Text price per 1M tokens	$0.25 input / $0.75 output	$0.75 input / $4.50 output
Audio price	Separate STT and TTS	$3 input / $12 output per 1M audio tokens
Stability	Hosted production API	Preview model

Google also publishes minute equivalents for audio: $0.005 per input minute and $0.018 per output minute. Those values make a first estimate easier, but accumulated context, thinking tokens, tools, and abandoned output can change the bill.

Mercury 2 is a text-system decision

Mercury 2 is a diffusion reasoning model with a 128K window, adjustable reasoning, tool use, schema-aligned JSON, and an OpenAI-compatible chat-completions endpoint. Inception lists $0.25 per million input tokens and $0.75 per million output tokens.

The current hosted row measures 802 output tokens per second and 3.53 seconds to the first answer. Inception's Blackwell test reports 1,009 tokens per second. The difference is not an error: prompt shape, hardware, load, and timing method differ.

For voice, Mercury sits between a speech recognizer and a speech generator. That architecture lets a team choose a specialist transcriber, preserve an existing voice, route regions independently, and replace Mercury later. It also means the team owns the seams and their latency.

Gemini Live is an audio-system decision

Gemini 3.1 Flash Live Preview connects through Google's Live API and processes audio directly. Google describes it as an audio-to-audio model optimized for realtime dialogue, acoustic nuance, numeric precision, and multimodal awareness.

The capabilities guide lists 16 kHz native input audio and 24 kHz output audio, with resampling available. Audio-only sessions can run longer than sessions that combine audio and video. Function calling is supported, and Google's current tool guide documents synchronous calls for the 3.1 model.

The model can preserve information that a transcript loses, including hesitation, pronunciation, timing, and tone. It can also accept image or video context. Those capabilities are valuable for visual support, field work, tutoring, and any call where “what I see” belongs beside “what I hear.”

The multimodal path changes evaluation. A field-support replay should include a blurry image, a camera that moves while the user speaks, a correction to a serial number, and a tool call grounded in both modalities. Score whether the cited visual detail was actually present, not only whether the response sounded confident.

Preview status is the counterweight. Model identifiers, rate limits, behavior, and availability can change before general release. A production plan needs a pinned acceptance suite and a migration path.

That path should define the replacement model, feature differences, a rollback threshold, and how active sessions drain during a change. Realtime systems make version drift immediately audible to users.

Keep model and API versions in every call trace so a behavioral change can be tied to the session that produced it.

Do not divide unlike latency numbers

Inception's voice report gives provider and customer figures for Mercury's reasoning and model layer. Google's Live documentation describes a low-latency endpoint but does not publish the same audio clip, region, tool call, load, and timing boundary.

Mercury's 802 tokens per second says nothing about speech recognition delay or the time until its text becomes audible. Gemini's native audio design removes those external model calls but still has turn detection, network, reasoning, synthesis, and playback time.

Use one clock: end of user speech to first audible agent sample. Replay identical audio into both systems. Track p50 and p95, then record false barge-ins, missed interruptions, tool completion, transcription-sensitive errors, and task success.

One sentence should expose the difference: “No, change the delivery address to sixty-eight, not sixteen, King Street.”

Compare the whole call bill

Mercury's model-layer cost is easy to compute from text. The rest of the chain is not included. Add speech recognition, speech generation, telephony, orchestration, tool calls, logging, and the cost of repeated context.

Gemini's audio rates provide a minute estimate, but Google's Live API guidance explains that accumulated session context can be processed again as a conversation grows. Thinking tokens are included in output pricing. Search grounding can generate one or more billable queries after the monthly allowance.

Capture real usage traces from the same replay set. A theoretical five-minute call with continuous audio can differ sharply from a call with long silence, interruptions, and one expensive search tool.

Choose the failure mode you can operate

Mercury 2 fits teams that want:

• independent speech recognition and voices,
• very low text-token prices,
• OpenAI-compatible text integration,
• schema-aligned tool output, and
• control over component-level routing.

Gemini 3.1 Flash Live Preview fits teams that want:

• native acoustic and spoken context,
• audio-to-audio response generation,
• image or video context in the live session,
• Google Search grounding and function calling, and
• fewer model boundaries in the conversational path.

Mercury asks the team to integrate more. Gemini asks the team to tolerate more platform coupling and preview risk.

That is the decision the benchmark must test.