llm-gemini 0.37 notes Gemini 4 Hydrogen launch and prices
Hydrogen is the cheapest Gemini 4 model at $1.50 per million input tokens; Carbon and Argon prices double December 1.
Hydrogen is the cheapest Gemini 4 model at $1.50 per million input tokens; Carbon and Argon prices double December 1.
The fast model costs $1.50 per million input tokens and is live in the Gemini app and API.
Google says Hydrogen is 2.5 times faster than Gemini 3.8 Flash and is rolling out to Search users globally.
Google says high mode is 2.5 times faster than 3.8 Flash and cuts fabricated claims more than fivefold.
Google prices the 1 million-token model at $1.50 per million input tokens and $7.50 per million output tokens.
Flat pricing is $1.50 per million input tokens and $7.50 per million output tokens across a 1 million-token window.
The company says annualized revenue reached $3 billion and it has 5 million experts.
Opus 5.5 and GPT-6 Astra each solved all 55; Opus found 72 new frontier points to Astra's 32.
Paid users can access Carbon now; Google says it complements Argon but spends more tokens on hard tasks.
DeepMind reports 96 percent success sorting parts and is rolling the model out to Agile, Apptronik and Mercedes-Benz.
Carbon uses Argon's introductory token price and, Google says, leads several coding and agent benchmarks at lower cost.
Introductory pricing matches Gemini 4 Argon at $2 per million input tokens and $10 per million output tokens.
Apollo 2 robots kitted parts in Mercedes factories at 96% success; the open jailbreak set has 1,500 prompts.
DeepMind reports 30% to 100% higher success, faster tasks, and 96% kitting success on Mercedes-Benz lines.
RoboRisk runs 92 simulated and 13 physical tasks; six foundation models generally showed high harm rates.
A judge trained on 3,788 videos tracks human ratings more closely than other open judges and matches Gemini 4 Argon.
New sandbox and VM isolation options let businesses verify CodeMender vulnerabilities while centralizing agent credentials via Auth Manager and Agent Identity JWT.
Gemini 4 Argon finishes 53.4% of 315 scenarios in time; a human study finds 78.5% metric agreement.
Gemini 4 Carbon pre-releases for government cyber defense while Genie 4 enables real-time simulations.
Selected defenders get first access; Google cites 88.2% on SEC-bench Pro and has not set a general-availability date.
Limited Fairwind access comes first; Google cites 88.2% on SEC-bench Pro and prices of $2 and $10 per million tokens.
Access is limited to some cyber defenders under Fairwind, and Google says it resisted every tested indirect prompt injection.
It scores 59 and is limited to selected users, trailing Claude Opus 5.5 by 3 points on the coding-agent index.
Carbon leads Artificial Analysis Intelligence Index v5.0 at 59 versus Argon’s 50, with stronger security-benchmark scores.
Google says Carbon tops Argon on its tests, including cleaner code, but the staged release remains closed.
Selected cyber defenders get access this week via Fairwind, priced at $2 and $10 per million tokens.
Limited access starts with cyber defenders under Fairwind, at Argon’s $2/$10 per million token prices.
Only select testers have it now; Google AI Ultra subscribers and early Gemini API partners are due next, with SynthID on outputs.
Trusted testers get it now; Google AI Ultra and Gemini API partners follow in the coming weeks.
A research preview goes to trusted testers first, then Google AI Ultra subscribers in Project Genie and early Gemini API partners.
Nvidia says three fine-tuned Qwen3.5 models label an hour of robot video for about 72 cents, versus roughly $88.50 with GPT-6 Astra.
The author says CVE-2026-109618 and CVE-2026-109619 were exploited before the delayed patch, after leaks broke the embargo.
ModelCompare says 29 of 35 answers claimed people put group belonging ahead of truth.
Calif, using GPT-5.6 Sol, and Wiz independently found CVE-2026-109618 and CVE-2026-109619 after about two months of private coordination.
OpenAI says the models found the issues before attackers, and a GitHub tool checks whether a PHP version is affected.
OpenAI and Wiz say the holes were found before attackers; unpatched sites should go offline, and PHP 8.5.12 is recommended.
Counting evidence sources blocked 0.9% of correct payments versus 7.4% for the best vote-counting rule, at equal safety.
They slew the Ender Dragon in 27 hours 14 minutes after writing 2,783 programs and dying 605 times.
AINews covers Meta’s Muse Spark Open and Reflection Beam 1.0 alongside Gemini, Anthropic, and OpenAI updates.
Gemini 4 Argon scored 97.5% on ARC-AGI-3, near GPT-6 Astra but weaker in math and coding.
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.
Today 121 readers · 129 views · 26 reads · all time 199K views