Breaking

Unreleased Gemini 3.8 Live Identifiers Surface as Musk Sets Grok 4.7 at Opus 5 Level

September 15, 2026 at 05:33 EDT

Foundation Models
AI Agents
Software Dev & Coding

Unreleased identifiers for Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking have surfaced on Google Cloud quota and metrics pages, pointing to a possible next step for Google's real-time Gemini lineup, but Google has not announced either model or published a public endpoint. The signal lands as Musk has set Grok 4.7 expectations around Claude Opus 5-level performance rather than Anthropic's newer Claude Fable 5.1, with multimodal performance still described as unfinished.

Sep. 15 · Model watch

Gemini Live names surface

Unreleased Gemini 3.8 Live identifiers appeared before a public endpoint, while Grok 4.7 expectations are being framed around Opus 5-level performance rather than the newer Fable 5.1 tier.

unreleased Gemini 3.8 Live names

1,048,576

Gemini 3.8 Flash context tokens

$0.75

Gemini 3.8 Flash input price per 1M tokens

500k

Grok 4.6 public context window

Gemini 3.8 Flash moves the public baseline

Terminal-bench 2.1 scores, shown as proportional stacked columns.

90.8%

Gemini 3.8 Flash

81.6%

Gemini 3.7 Flash

The public 3.8 model is stronger on this coding-agent benchmark, but it is not a Live API model.

Why builders care

A 3.8-class Live model would matter for low-latency voice, video assistants, tutoring, games and other real-time agent interfaces.

Why teams wait

There is no public Gemini 3.8 Live endpoint, quota, price or model card yet, and Gemini 3.8 Flash itself has no Live API support.

What would make it real

Public endpoint

→

Model card and pricing

→

Callable API access

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.