“AI Disruption” Publication 10,500 Subscriptions 20% Discount Offer Link.
Google(GOOG 0.00%↑) is back.
Today, Google DeepMind dropped two models at once: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. Officially, they are Google’s most advanced real-time conversation models to date.
The global end-to-end speech leaderboard has a new champion: Gemini 3.8 Live Extended Thinking (High) took the top spot with a composite score of 82.6, knocking off OpenAI’s GPT-Live-1 (Astra). Then came the price cut. The standard version costs just $0.84 per hour of audio—the cheapest anywhere. The top thinking tier is $3.50, about one-third the price of OpenAI’s realtime API.
One caveat in the chart: they compared Gemini’s High tier against OpenAI Astra’s Medium tier. The reason is practical. OpenAI’s high-thinking realtime API is extremely expensive and tightly rate-limited, so evaluators could only compare against same-price or currently stable public tiers. Gemini still beat the competitor’s top model at a lower price. The compute-to-performance ratio is already visible.
Gemini 3.8 Live Standard: $0.84 per hour of input audio, slashing the previous generation and becoming the cheapest commercial real-time speech model in the industry.
Gemini 3.8 Live top thinking tier: $3.50 per hour—40%–60% cheaper than OpenAI (Sol $4.47, Astra $5.83), and less than one-third of the old GPT-Realtime 2.1 at $10.75/hour.
