Recent stories
Each story's summary, with the places it was discussed.
Summaries are written by AI from the linked sources. Open the source for full context.
SpaceX has entered a $920 million monthly deal with Google to provide compute resources including NVIDIA GPU and CPU hardware.
This agreement, preceding SpaceX’s historic IPO, follows a similar partnership with Anthropic. Google intends to use this capacity to bridge demand for its Gemini Enterprise platform as Alphabet accelerates its massive capital expenditure strategy.
Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5
Artificial Intelligence · Large Language Models · Gemini · LLM Benchmarks
Google’s Gemini 4 Carbon ranks #1 on the Artificial Analysis Intelligence Index v5, scoring 59 and leading four of nine evaluations, while ranking second behind Claude Opus 5.5 on the Coding Agent Index.
It improves coding and agentic knowledge work over Gemini 4 Argon, but does not lead those areas. Carbon costs $7.76 per Intelligence Index task, uses a 1M-token context window, supports multimodal input, and is available only to selected users with a 50% launch discount.
Gemini 4 Neon looks like an Opus 5.5 competitor
Artificial Intelligence · Data Science · Large Language Models · Gemini · LLM Benchmarks
Gemini 4 Neon appears to rival Opus 5.5 by combining Gemini 4 Carbon-level performance with much higher speed.
Gemini 4 Carbon leads the Agent Duel leaderboard excluding early votes but generates 46 tokens per second, versus Neon’s reported 127. A model of leaderboard win rates found speed nearly as important as model identity. Assuming Neon matches Carbon’s performance, it projects to 1,066 Elo, just below Opus 5.5’s 1,078, with a 48.3% win rate against it.
Google has rebranded NotebookLM as Gemini Notebook, enhancing it with native code execution for complex data analysis.
Available to premium users, this research tool now integrates more deeply across the Gemini app and Google Search, allowing for cross-platform synchronization and improved data processing capabilities grounded in user-provided sources.
Google is rolling out Gemini Omni 1.2 Flash generally across the Gemini API and Gemini for business, adding native video creation and editing up to 15 seconds, stronger reference adherence, improved consistency, natural speech, and more controllable camera motion.
The model costs $0.07 per second for 720p video, 30% less than Omni 1.1, supports conversational edits and upscaling to 1080p or 4K, and includes enterprise security, Provisioned Throughput, and SynthID watermarking.
This technique enables reliable text and number generation in AI images by using a two-layer approach.
First, an 'underdrawing' layer is created deterministically with precise layout constraints. Second, this base image is passed to a multimodal model like Gemini 3.0 Pro, which uses the structure to paint the final generative image accurately.
The open-source /debuzz skill for Claude Code pipes Claude’s responses through Gemini CLI to remove theatrical, clickbait-like language while preserving technical details.
It supports colleague, manager, and director modes, can process previous replies or pasted text, and prints Gemini’s translation verbatim, with a labeled fallback if Gemini fails. Licensed MIT.
Gemini 4 Carbon is now available
Artificial Intelligence · Large Language Models · Gemini · AI Agents
Google has made Gemini 4 Carbon available to developers, enterprises, scientists, and consumers as its strongest model for cybersecurity, long-horizon agentic workflows, complex coding, knowledge work, and scientific reasoning.
Carbon ranks first on CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench at over twice the cost efficiency of the next external model. It supports coordinated /swarm agents in Google Antigravity, costs $2 per million input tokens and $10 per million output tokens, and includes strengthened safety measures.
Google introduces Gemini 3.5 Transcribe, a speech-to-text model for polished, context-aware transcription.
It supports real-time streaming and pre-recorded audio APIs, custom vocabulary, 85+ languages, speaker attribution, timestamps, function calling, and noisy environments. It reports lower Word Error Rates and 70% faster final transcription than Chirp 3, powering voice agents, dictation, captions, analytics, and Gemini products.
Google introduced Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, enabling prompt-based voice creation, voice replication with consent verification, line-by-line performance direction, two-speaker scenes, long-form stability, and multilingual generation.
Flash targets creative control; Flash-Lite targets cost-efficient scale. SynthID watermarking, C2PA credentials, Google AI Studio, and Gemini API support safer deployment.
Google’s Gemini 3.8 Flash, based on Gemini 3.7 Flash, improves software engineering and agentic knowledge workflows with adjustable effort levels.
It accepts text, images, audio, and video, supports a 1M-token context window and 64K-token outputs, and is distributed across Google channels. Evaluations show comparable safety, slight non-English regression, and persistent hallucination, latency, and knowledge-cutoff limitations.
Gemini 4 Argon is introduced as a general-purpose physical-reasoning model for robotics, combining 3D understanding, code generation, visual-semantic reasoning, and control.
It scores 87% on ERQA, outperforming every other frontier model, and supports robot control, Real2Sim reconstruction, physical understanding, and research/self-improvement. Examples include MuJoCo twins, zero-shot manipulation, hindsight labels, and browser-based paper reproductions. In simulation, 4,500+ episodes raised real-robot Success Rate from 56% to 75% and cut completion time from 35s to 24s. It is available via Google AI Studio, Gemini API, and Google Antigravity.
Gemini Omni 1.1 Flash adds controllable generative video features through Gemini API and Google AI Studio.
Developers can extend scenes using 10 seconds of context up to 40 seconds, define first and last frames, use video references, draft in faster, cheaper 360p, and upscale outputs to 1080p or 4K for production workflows.
Google introduced Gemini Omni, a multimodal AI model designed for realistic video generation and creative editing.
By integrating physics, cultural context, and advanced reasoning, it enables cohesive scene construction. The release emphasizes safety through rigorous red teaming and responsible AI principles, while incorporating SynthID and C2PA watermarks to ensure transparency in AI-generated content across digital platforms.
Gemini 4 Carbon: Our next frontier model
Artificial Intelligence · Cybersecurity · Large Language Models · Gemini · LLM Benchmarks
Gemini 4 Carbon is a new frontier model focused on long-horizon engineering and cybersecurity, ranking #1 on the Artificial Analysis Intelligence Index v5.0 with a score of 59.
It leads SEC-bench Pro (88.2%), ExploitGym (54.8%), and CWE-bench (71.0%), while improving secure-code generation, alignment, honesty, and prompt-injection resistance. Carbon can sustain agentic workflows for days. It will initially reach trusted cyber defenders through the Fairwind Program at Argon’s introductory price.
Developing Gemini 4 Argon responsibly and safely
Artificial Intelligence · Cybersecurity · Large Language Models · Ethics · Gemini
Google is beginning broad rollout of Gemini 4 Argon after limited testing with Fairwind cyber defenders and security teams.
Argon found a critical bug in hospital software missed by previous models and solved all 12 simulated end-to-end cyberattack scenarios, versus five previously. Its safeguards use input screening, activation probes, continuous monitoring, and an automated misalignment monitor. Bio jailbreaks unlocked under 1% of tested sub-tasks, shortcut attempts fell to 2%, and Indirect Prompt Injection attack success reached 0.7%.
Google to shut off search for schools that opt-out of AI
Artificial Intelligence · News · Education · Ethics · Gemini
Google will disable Gemini, Gemini Notebook, Gemini in Classroom and Google Search by default for K-12 student accounts, requiring administrators to activate them.
After backlash, it clarified that schools can enable Search separately, though AI features such as AI Overviews and AI Mode remain integrated. Google says Search will guide students through problems rather than provide direct answers and detect answer-seeking use. Critics call the policy pressure to adopt Google’s AI amid privacy and learning concerns.
Gemini 3.5 Flash now includes built-in computer use capabilities, allowing developers to create agents that interact with browser, mobile, and desktop environments.
This integration enhances performance for enterprise automation and software testing. Google also introduced safety features, including adversarial training and optional enterprise safeguards, to mitigate risks like prompt injection.
Google is evolving the mouse pointer for the AI era by integrating Gemini to provide context-aware assistance.
By prioritizing flow, visual cues, and natural language, the new AI-enabled pointer understands screen content beyond just coordinates, allowing users to perform complex tasks through simple gestures and speech across all applications without interrupting their workflow.
Google launched Gemini 4 Neon, a frontier-level model for complex coding, analytical, scientific, enterprise, and cybersecurity workflows.
It offers a 1-million-token context window, four thinking levels, and pricing of $3 per million input tokens and $15 per million output tokens. Neon ranks highly across coding and agent benchmarks, including #1 on Artificial Analysis Coding Agent Index, while emphasizing lower task costs, multi-step execution, cyber defense, prompt-injection resistance, and reduced fabrication. It is rolling out through Gemini API, Google Antigravity, Gemini for business, and the Gemini app.