Image to help understand the article
Google unveiled Gemini 3.8 Live with Live Avatar on Sept. 24, 2026, introducing a system that processes voice and video in real time and synchronizes multilingual speech across 97 languages.
The company said content generated by the artificial intelligence system will carry its SynthID watermark, Google’s technology for identifying AI-generated media.
Enterprise Release in the US and European Union
Google said Gemini 3.8 Live with Live Avatar is available through Gemini Enterprise. Google Cloud announced Sept. 25 that the feature had reached general availability through the Gemini Enterprise console and Gemini Live API, using endpoints in the United States and European Union. Customers can also purchase provisioned throughput, which reserves processing capacity for more predictable performance.
Custom avatar creation remains limited to corporate customers on an approved access list. Android Headlines reported that the technology was previewed at Google Cloud Next 2026 before its commercial release with enterprise compliance assurances.
Google Cloud identifies the model as Gemini 3.8 Live but has not listed a separate general-release model ID for the avatar feature. Gemini 3.8 Live Extended Thinking remains in private preview. DataCamp lists the model IDs as gemini-3.8-live and gemini-3.8-live-extended-thinking.
Usage-Based Pricing
Google Cloud’s announcement did not include specific prices, instead directing customers to separate pricing pages for Gemini Enterprise and Agent Platform. DataCamp reported rates per 1 million tokens of $3 for audio input, $12 for audio output, 75 cents for text input and $4.50 for text output. The audio output rate includes tokens used for the model’s internal reasoning.
Google Search grounding, which allows the AI to draw on search results, includes 5,000 requests per month at no charge, followed by a rate of $14 per 1,000 requests, according to DataCamp.
Independent Tests Show Mixed Results
Google’s announcement did not provide benchmarks or direct comparisons with competing models. Independent testing by Artificial Analysis produced varying results for the standard and Extended Thinking versions.
On the firm’s Speech to Speech Index, Gemini 3.8 Live Extended Thinking at its High setting scored 82.6, with 68.6% for agentic performance, an Arena Preference Elo rating of 990 and an 89.1% task success rate. Standard Gemini 3.8 Live scored 76.0, 30.1%, 1083 and 93.2%, respectively. OpenAI’s GPT-Live-1 at its Sol low setting recorded a 90.9% task success rate.
DataCamp’s reported τ-Voice results placed Gemini 3.8 Live Extended Thinking at 68.6%, narrowly ahead of GPT-Live-1 Astra at 67.9% and Grok Voice Think Fast 2.0 at 56.5%. Their Speech to Speech Index scores were 82.6, 81.5 and 81.3, respectively.
Other tests favored rival products. Artificial Analysis data cited by BeInCrypto showed Qwen Audio 3.0 scoring 99.2% on a speech-reasoning benchmark, compared with 97.7% for Gemini 3.8 Live Extended Thinking.
Human preference findings also differed by source. Artificial Analysis said standard Gemini 3.8 Live had the highest Arena Preference Elo among the models it evaluated, at 1083. BeInCrypto, citing Artificial Analysis data, reported that the earlier Gemini 3.1 Flash Live scored 1096, higher than both Gemini 3.8 Live versions. The Extended Thinking model’s reported rating was 990.
Limits and Competitive Pressure
Memeburn cited a separate software-engineering benchmark in which Claude Sonnet 4.6 scored 82.1% and Gemini 3.1 Pro scored 63.8%. That comparison did not involve Gemini 3.8 Live, and the publication noted that Google had not released a SWE-bench score for its latest model.
Memeburn described Gemini as the only major model to accept video and audio as native input formats, saying Claude and GPT-5 did not offer those inputs. DataCamp cautioned that the Gemini 3.8 Live family can still produce false information, remains under development in its resistance to jailbreak attacks and may experience occasional slowdowns or timeouts.
Android Headlines characterized Live Avatar as a response to intensifying competition rather than a stand-alone breakthrough. The publication said Google shares had fallen about 20% in recent months following reports of delays involving Gemini 3.5 Pro.
Comments
Post a Comment