Models & Labs
Google deploys Gemini 4 Hydrogen for faster Search
TechCrunch AIhigh confidence
Why it matters
- →Google prioritizes speed and cost-efficiency over raw model size for high-volume search tasks.
- →The integration of specialized fast models into consumer search improves latency for complex, multi-step queries.
- →Hydrogen's ability to outperform larger models on specific benchmarks highlights the value of task-specific optimization.
Google has integrated its new Gemini 4 Hydrogen model into Google Search globally, aiming to improve response speed and accuracy for complex queries. Designed as a fast-execution model, Hydrogen is reported to be 2.5 times faster than the previous Gemini 3.8 Flash while maintaining high intelligence levels, even outperforming the more powerful Gemini 4 Argon on certain benchmarks. The update enhances Search's 'fan-out' technique, allowing for better handling of multi-part questions and typos, and introduces richer visual responses like comparison tables in AI Mode. Hydrogen is also available via the Gemini API and Studio, offering four thinking levels to balance performance and token usage.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1Google Revamps Gemini AI Usage and Pricing
WIRED AI · July 18, 2026 · Related
- 2Google Launches Gemini 3.8 Flash Model
The Verge AI · September 2, 2026 · Related
- 3Google Announces Gemini 4 Argon with 1M Token Output
Sam Witteveen · October 1, 2026 · Related
- 4Google Gemini 4 Argon Benchmarks and Coding Performance
The AI Daily Brief · October 14, 2026 · Related
- 5Google Gemini 4 Carbon wins benchmarks
Matthew Berman · October 29, 2026 · Same story
- 6Google Releases Gemini 4 Carbon for Cost-Efficient Inference
Matt Wolfe · October 30, 2026 · Same story
- 7Google launches Gemini 4 Carbon with agent swarm
The Verge AI · November 10, 2026 · Related
- 8Google Releases Gemini 4 Carbon Model
The AI Daily Brief · November 12, 2026 · Related
Follow this story
Open the full story →Google deploys Gemini 4 Hydrogen for faster Search
2 developments
- Nov 12 · TechCrunch AI
Google deploys Gemini 4 Hydrogen for faster Search (This article) - Nov 12 · The Verge AI
Google launches Gemini 4 Hydrogen for complex queries
More from TechCrunch AI
© TechCrunch AIAgentsagents
Meta Muse AI Agent Now Places US Business Calls
Meta’s consumer agent Muse is rolling out voice calling to US businesses for paid subscribers, targeting the tedious task of customer service navigation. With over 10 million users, Meta is betting that AI’s infinite patience will replace human frustration with phone trees and hold times. The feature leverages improved voice mimicry to interact with receptionists, though many still hang up upon detecting non-human voices. This moves Muse from text-based assistance into active, autonomous execution of real-world tasks, marking a significant step toward mainstream consumer AI utility.
TechCrunch AI
© TechCrunch AIMarket & Regulationbusiness
Nebius delays data centers, cuts revenue forecast
Nebius is facing the harsh reality of scaling AI infrastructure: even with $40 billion in planned spending and massive prepayments from Microsoft and Meta, physical construction delays are derailing timelines. A key New Jersey site is months behind schedule due to regulatory fines and generator issues, forcing a shift in delivery dates to 2027. This isn't just a logistical hiccup; it signals that the 'neocloud' model of financing builds with customer cash is fragile when third-party operators fail. The market reacted swiftly, dropping shares by 17% after Nebius cut its 2026 revenue forecast, proving that hype cannot override concrete capacity.
TechCrunch AI
© TechCrunch AIAgentsagents
Nectar launches AI agent for influencer outreach
Nectar’s Creator Agent automates the messy middle of influencer marketing by scanning 300 million profiles to find and vet creators. It goes beyond keyword matching, analyzing video content for brand safety and context rather than relying solely on keywords. The tool drafts personalized messages in a brand's voice, manages contracts, and tracks campaign performance on Instagram and TikTok. This marks a shift toward agentic workflows in marketing ops, though it competes with established players like Aspire. By handling logistics from contracts to gifting, it aims to replace manual spreadsheet management with autonomous relationship building. The system adapts tone for different social environments, ensuring brands sound appropriate whether posting on X or commenting on TikTok. This approach reduces the friction of scaling creator partnerships without losing human nuance.
TechCrunch AI
More in Models & Labs
Models & Labsother
llama.cpp b12340 adds CUDA 13 and ROCm 10.0
This release quietly closes the hardware gap for local inference by adding default support for CUDA 13 and ROCm 10.0 across Linux and Windows. Users with newer NVIDIA GPUs or AMD accelerators no longer need to compile from source to get performance, while Apple Silicon builds now ship KleidiAI disabled by default, shifting the focus back to standard ARM optimizations. The inclusion of Snapdragon support for Linux and Android marks a significant step toward on-device AI beyond the x86/ARM desktop ecosystem. It is less about new features and more about ensuring llama.cpp remains the universal runtime for every available accelerator.
llama.cpp Releases
Models & Labscoding
llama.cpp b12342 adds few-row MMQ for speculative decoding
Speculative decoding just got a significant speed boost on NVIDIA GPUs. This release introduces a specialized kernel for 'few-row' matrix multiplications, targeting the small batch sizes typical of verification passes in speculative decoding workflows. By optimizing these specific operations for quantized formats like Q8_0 and IQ4 variants, llama.cpp reduces latency where it matters most during token generation. The change is CUDA-only for now but leverages hardware capabilities from Ampere onwards, making local inference faster without requiring new model architectures or complex configuration changes.
llama.cpp Releases
© Hugging Face BlogModels & Labscoding
IBM releases LLMON markup to fix prompt injection
IBM is tackling the brittleness of small language models by introducing LLMON, a structured markup language that explicitly separates instructions from data. By using reserved special tokens instead of ambiguous characters like JSON brackets, these new Granite 4.1 preview models drastically reduce indirect prompt injection attacks while maintaining standard performance. This approach forces the model to bind specific constraints to specific data payloads, eliminating the confusion that often leads to hallucinations or security breaches in long contexts.
Hugging Face Blog