Bitcoin (BTC) Trades Near $76K as Google Launches Gemini 3.8 Live Audio Models

Benchmarks Rank the New Voice Models

Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, billing the pair as its most advanced audio models yet, and the launch landed on a crypto tape where Bitcoin (BTC) held near $76,000 at press time. The company confirmed the rollout in an official post on X, describing production-ready voice models that can speak, reason and carry out tasks in conversation. Independent scoring followed almost immediately — and it put the new high-effort variant at the front of the field. On Artificial Analysis's Speech-to-Speech Index, which blends speech reasoning, agentic performance, arena preference and task-success rates, Gemini 3.8 Live Extended Thinking, tested at high reasoning effort, debuted in first place at 82.6. GPT-Live-1 Astra followed at 81.5, with Grok Voice Think Fast 2.0 High at 81.3. The standard Gemini 3.8 Live placed fifth at 76.0, and both new variants outscored the previous flagship, Gemini 3.1 Flash Live High, which managed 71.5.

official post on Xhttps://x.com/Google/status/2099907828328595675?ref_src=twsrc%5Etfw

The agentic jump is the sharpest single improvement. Extended Thinking reached 68.6% on the Tau Voice benchmark against 37.7% for the prior generation — a 30.9-point step up in one release cycle. On speech reasoning it scored 97.7%, edging Grok Voice at 97.2% but still trailing Qwen Audio 3.0 Realtime Plus at 99.2%. For crypto readers, the release is another datapoint in the AI-compute buildout: every capability jump at hyperscaler scale feeds demand for frontier-grade memory and compute, the supply chain where names like Micron Technology sit. Bitcoin's proof of work network does not draw on that capex chain directly, but trader attention has historically rotated between AI and digital-asset risk, and market reaction to AI milestones is watched accordingly.

Pricing, Latency and the Preference Gap

Where the newest models separate from the field is price and speed. The standard Gemini 3.8 Live runs at $0.84 per hour of input audio — roughly half its predecessor's $1.75 rate and the cheapest price in the entire Index. Extended Thinking costs $3.50 per hour, undercutting GPT-Live-1 Sol at $4.47 and Grok Voice Think Fast 2.0 High at $4.80; a developer choosing deep reasoning over fast conversation therefore pays more than four times per audio hour. Latency improved in parallel: time to first audio averaged 1.18 seconds, down from 2.99 seconds for the prior Gemini model. Human preference, however, tells a different story. In blind Speech Agent Arena conversations, the older Gemini 3.1 Flash Live still leads outright with an Elo of 1096. The new Gemini 3.8 Live sits second at 1083, while Extended Thinking trails at 990 — despite completing 89.1% of its assigned tasks. The pattern suggests listeners still prefer the older model's conversational feel even as the newer variants win on measurable reasoning and task completion. Voice AI is now graded on two scales pointing in different directions: benchmark scores reward the model that reasons hardest, while arena preference rewards the one that talks best. Google currently tops both scales — with two different models — a dual lead no rival can claim at present, a structure that will shape how buyers in this market price quality, much as digital-asset desks separate hard settlement metrics from softer attention signals in memecoin trading, two gauges that can diverge for long stretches before converging. Readers tracking the market in real time can follow live spot and futures prices on Bitget.

AI Momentum Meets a Quiet Crypto Tape

Read together, the release and its reception sketch one arc: Google now leads voice AI on both measurable capability and — with its older model — human preference, while the crypto market itself generated little native newsflow for the session. Our desk treats the launch as confirmed on the primary record: Google's own announcement post states both model names and the September 15 date, and the rankings come from Artificial Analysis's published index rather than vendor self-reporting. As of the latest COINOTAG snapshot, Bitcoin spot trades at $75,854, close enough to round to the $76K headline level. The open question for digital assets is whether AI-compute momentum keeps absorbing risk capital that might otherwise chase spot Bitcoin ETF exposure — a rotation metric we will keep tracking.