Stats Wire

I can officially retire from X now. Sam Altman just quoted my tweet. 🔥

Gemini 4 Carbon High is SURPRISINGLY token-efficient on Google AI Ultra!!! I've been using Carbon High and Argon High heavily since my weekly quota reset. And look at my usage now: • Gemini weekly limit: 49% LEFT • 5-hour limit: 95% LEFT • Weekly reset: Still 1 day and 19 hours away!!! I've been throwing some pretty demanding coding tasks at Carbon, and honestly, I expected my quota to disappear MUCH faster. Google might have done something REALLY good with Carbon's usage optimization. Anyone else noticing this???

Gemini 4 Hydrogen's FULL pricing just surfaced!!! • Standard: $1.50 input / $7.50 output • Batch/Flex: $0.75 input / $3.75 output • Cached input: Just $0.075 Now compare standard INPUT pricing: • Gemini 4 Hydrogen: $1.50 • Claude Opus 5.5: $4 • Claude Fable 5.2: $8 • GPT-6 Astra: $10

CONFIRMED: Gemini 4 Hydrogen is dropping today.

DeepSeek V4.1 Pro just surfaced in Vercel's source code!!! • DeepSeek V4.1 Flash: Listed in the public catalogue • DeepSeek V4.1 Pro: Found in a hard-coded model-picker test Both identifiers appear in Vercel FX's model catalogue code.

HOLY SHIT, Gemini 4 Hydrogen is showing up in Google Cloud's PUBLIC billing catalogue!!! And the pricing is INSANE. Per 1M tokens: • Standard input: $1.50 • Cached input: $0.075 • Batch input: $0.75 • Batch output: $3.75 • Flex output: $3.75 Hydrogen is: • 2.7x CHEAPER than Opus • 5.3x CHEAPER than Fable • 6.7x CHEAPER than Astra The catalogue contains 69 Hydrogen billing SKUs!!!

Google is COOKING something with Antigravity!!! I've been getting update after update over the last few hours. Every time I open Antigravity, there's another "Restart to Update" button waiting for me!!! Google, WHAT ARE YOU COOKING???

Gemini 4 Hydrogen is now on Vertex AI' • Model ID: gemini-4-hydrogen (promoted from -eap) • modelVersion: "gemini-4-hydrogen" • Self-ID: "I am Gemini 4 Hydrogen, a large language model built by Google."

Gemini Hydrogen dropping TODAY 🎉 When Logan says Gemini, new model is about to drop

Google is preparing an XHIGH reasoning tier for Gemini!!! And this time, it's showing up in Google's own GenAI SDK. 'ThinkingLevel.XHIGH' Meanwhile, Gemini 4 Carbon currently offers: • Low • Medium • High

Gemini Hydrogen gets fresh corroboration: today's Google sync again reports gemini-4-hydrogen-eap among API-returned IDs. This supports the existing EAP lead.

Gemini 4 Carbon High just worked on my task for 1 HOUR and 20 MINUTES!!! And then THIS happened: "Deadline exceeded: Agent execution terminated due to error." 80 MINUTES of execution, only to end with a timeout!!! And this isn't the first time. I've faced this TWICE since Carbon launched yesterday. Before Carbon's release, I NEVER encountered this issue in Antigravity. Anyone else experiencing this since yesterday???

Google is preparing to launch a new model on Thursday/Friday It's just Wednesday Team is cooking

Gemini Decision 1 just dropped its first public numbers. It leads most of the key benchmarks: JevBench 92.2% Amazon ESCI 58.7% New Yorker Caption 79.7% PubMedQA 79.8% Global-MMLU-Lite 86.0% Strongest calibration too (lowest Top-1 ECE at 0.0250). TypeSafe is cheaper and wins a couple of categories. But on pure decision quality, Google’s new model is currently ahead.

Google employees are ALREADY testing Gemini 4 Hydrogen!!! • Codename: Hydrogen • Focus: Speed and efficiency • Internal access: Google's Jetski platform • Expected positioning: Next-generation Flash A Google engineer described it as "very capable, fast and efficient with a much FASTER and cheaper Flash alternative"

Here you go

Gemini 4 Carbon is available in Google Search AI Mode too!!! Just tested Google Search, and look what appeared in the model selector: • Auto • Gemini 3.8 Flash • Gemini 4 Carbon Carbon is now showing up directly inside Google Search!!! If you're on Google AI Pro or even the FREE plan, go check your AI Mode model selector RIGHT NOW.

Google is preparing an unannounced Gemini Decisions API its answer to OpenAI's Decisions endpoint. Spotted in GenAI SDK PR #3353 (docs still 404): • decisions-prerelease • POST /v1beta/decisions • Fast batch yes/no probabilities, multiple-choice & rubric scoring

Biggest week of November so far. Monday: Claude Fable 5.2 Tuesday: Gemini 4 Carbon Wednesday: Gemini 4 Hydrogen Thursday: GPT-6.2 Astra Four frontier models in four days.