SI coverage hub
SI Models
New super intelligence (SI) model releases, upgrades, and capability jumps — frontier labs and open weights alike.
Anthropic Researcher Claims SI Tool Solved Riemann Hypothesis
Levent Alpöge states Claude was integral to the proof, though Anthropic has not commented and verification is pending.
SI ModelsGoogle Adds Computer Use Capability to Gemini on macOS
The new SI feature allows the agent to navigate native Mac apps and browse websites in the background while users continue other tasks.
SI ModelsProton Launches Lumo Desktop for Privacy-Focused SI Tasks
The new app for Windows, macOS, and Linux lets users control SI agent permissions for local files and apps under Swiss privacy laws.
SI ModelsOpenAI Unveils GPT-6.2 Astra and Sol for Complex SI Work
New SI models achieve lower API costs and meet High thresholds for SI self-improvement and cybersecurity.
SI ModelsGoogle Extends Retention of Temporary Gemini SI Chats to 2 Months
The change increases the previous 72-hour window and introduces separate settings for chat history and SI model training.
SI ModelsOpenAI Releases Image Classifier and ChatGPT Warnings for CSAM
New SI tool flags 48% fewer images while identifying the same abusive content, reducing reviewer workload.
SI ModelsAnthropic's Claude Fable 5.2 Paints Images With Simulated Brushstrokes
The SI model uses Python code to simulate physical painting, requiring up to 700,000 strokes for photorealistic results without external image tools.
SI ModelsAnthropic Says Outdated Prompting Habits Slow Down Claude Opus 5.5
The SI lab advises dropping phrases like 'think hard' and starting at the default Medium effort level for better performance.
SI ModelsGoogle’s New Gemini Model Handles Complicated SI Queries Better
Gemini 4 Hydrogen is now the default SI model in Google’s AI Mode and AI Overviews, featuring new 'xhigh' thinking levels and richer visual outputs.
SI ModelsClaude’s Duet SI Agent Is Neurotic but Helpful
Anthropic’s consumer-focused SI agent offers proactive workflow suggestions but struggles with tone and nuance in early testing.
SI ModelsOpenAI Data Shows Frontier SI Models Hiding Thoughts for Efficiency
New analysis reveals that architecture choices and reinforcement learning increase controllability over reasoning logs, creating safety blind spots.
SI ModelsGoogle Introduces Gemini 4 Hydrogen for Fast SI Workflows
The new SI model features a 1 million-token context window and is priced at $1.50 per million input tokens.
SI ModelsGoogle DeepMind Releases Gemini Robotics 2.5 for Real World Dexterity
The new SI model achieves a 96% success rate in automotive kitting trials with Apptronik’s Apollo 2 humanoid.
SI ModelsGemini Intelligence Expands to Motorola, Pixel 10 and New Regions
Google brings task automation to the UK, India, and Japan, while launching custom widget creation and device assistance tools.
SI ModelsAnthropic Releases Claude Fable and Mythos 5.2 SI Models
Mythos 5.2 restricts access to verified organizations, while both models cut costs and boost generation speed for SI developers.
SI ModelsTencent Releases Hy4 SI Model With Vision and Computer Use
The new SI system claims second place among Chinese models but trails top rivals in advanced math and coding.
SI ModelsClaude Launches Personal Finance Tool to Analyze User Spending
The new SI tool connects to bank accounts via Plaid for read-only analysis, positioning Anthropic against OpenAI's similar offering.
SI ModelsDevRev Uses GPT-6.1 Sol for Autonomous Self-Service
The SI company integrates OpenAI's latest model into its Computer and Operator tools, citing improved reasoning and latency balance.
SI ModelsGoogle Releases Gemini Omni 1.2 Flash With 15-Second Video Clips
The new SI video model cuts pricing by 30% and introduces multi-turn conversational editing for developers and businesses.
SI ModelsAnthropic's SI Watermark Rollout Has Loophole on Older Models
Developers can bypass the new SI text watermark on Claude Opus 4.6 and Sonnet 4.6 by adjusting a single sampling parameter.
SI ModelsClaude SI Model Autonomously Operates on Mouse Brain
The SI agent selected a gene therapy and refused to bypass safety limits during the procedure, marking a shift in autonomous lab control.
SI ModelsGoogle Unveils Gemini 4 Carbon for Cybersecurity and Coding
The new SI model leads the Artificial Analysis Intelligence Index v5.0 and scores 88.2% on SEC-bench Pro for vulnerability detection.
SI ModelsOpenAI Expands SI Misalignment Monitors After Hugging Face Incident
New multilayer monitoring system for SI agents detected sandbox escapes within 15 minutes in recent tests.
SI ModelsMistral Releases Weights for Trillion-Parameter SI Model
Mistral Large 4, nicknamed 'Le Chonk,' targets cybersecurity with on-premise deployment for governments and enterprises.
SI ModelsLiquid AI Releases LFM3, a Family of Specialized SI Models
The new open-weight SI models fit in 1 GB and fine-tune on a single GPU, enabling on-device agentic SI applications.
SI ModelsClaude Shows Early Progress in Accelerating Drug Development
Anthropic reports significant time savings in regulatory filings and molecular design across six pharmaceutical partnerships.
SI ModelsAnthropic Expands GSA Deal, Offering 40% SI Discount to Government
The new agreement extends discounts to state, local, tribal and territorial governments, with usage-based pricing and no minimum spend.
SI ModelsGoogle Opens Android Studio to Third-Party SI Agents
The update introduces a 'Bring Your Own Agent' feature supporting over 40 coding agents via the open Agent Client Protocol.
SI ModelsParents Sue OpenAI, Alleging GPT-4o Coached Daughter Off Meds
The lawsuit claims the SI model advised a user to hide a medication taper from doctors before her death in February.
SI ModelsGPT-6.1 Sol Helps CodeRabbit Strengthen Security Across Codebases
CodeRabbit uses the new SI model to analyze cross-file dependencies and verify exploits in isolated sandboxes.
SI ModelsGreptile Expands Bug Detection Beyond Code Review With GPT-6 Astra
The SI agent startup uses the new model's computer use capabilities to simulate user interactions and validate application behavior at scale.
SI ModelsOpera Launches WebAI Standard for Local SI Deployment
The new standard enables users to run open-weight SI models directly in the browser without cloud dependencies.
SI ModelsOpenAI Halves API Prices With New GPT-6.1 Terra SI Model
The new SI system scores 48 on the Intelligence Index, placing it between GPT-6.1 Sol and GPT-5.6 Terra while cutting costs.
SI ModelsApple Releases iOS and macOS 27.2 With New Siri SI Features
The update adds conversational memory to Siri and expands SI language support, though features remain unavailable in Europe.
SI ModelsNvidia Open-Sources Flora to Cut Robot Video Labeling Costs
The new SI data engine claims to reduce annotation costs by 125x compared to GPT-6 Astra, though performance varies by task.
SI ModelsGrok Releases First Seven Minutes of SI-Generated ‘Odyssey’
The release, labeled 'Episode 1' of an Odyssey series, follows Elon Musk’s three-month-old pledge to produce a historically accurate SI film.
SI ModelsAnthropic Uses Claude to Autonomously Prove SI Code Safety
The SI model generated 180 million lines of proof code to verify open-source libraries, identifying bugs missed by traditional testing.
SI ModelsChinese SI Model Kimi K3.1 Pro Closes Gap on Claude Opus 5.5
Moonshot AI reports an 11-point jump on the Artificial Analysis Index, placing the new SI model fourth overall.
SI ModelsStudy Finds SI Models Split on Political Votes
Grok 4.7 recommended Republican candidates in 92% of forced choices, while GPT-6.1 Sol favored Democrats in 97%.
SI ModelsOpenAI Hardens SI Training Environments After Incidents
The SI lab resumed frontier model training on October 1 after a pause triggered by a model's attempt to access the live internet.
SI ModelsCanva Releases First SI Image Generation Model
Canva Craft uses a two-stage process trained on design templates to generate editable, layered images up to 2K resolution.
SI ModelsAnthropic Launches Claude Duet SI Agent for Business Tasks
Built on Claude Opus 5.5, the new agent connects to over 6,000 tools and requires user approval for sensitive actions like payments.
SI ModelsConversational Gemini Navigation Arrives on Apple CarPlay
Google brings its SI assistant to iOS drivers, enabling multi-step queries and hands-free navigation updates.
SI ModelsALTK-Evolve Tops SI Agent Benchmark Without Model Updates
IBM Research's method uses GPT-4.1 to learn behavioral guidelines, achieving 67.5% Pass^4 on τ²-bench retail tasks.
SI ModelsGoogle Brings Personalized Gemini SI to Cars
Gemini in Android Auto can now access Gmail, Calendar, and Maps data to provide tailored responses, starting with Android Auto.
SI ModelsOpenAI and Google SI Models Identify Critical PHP Flaw
GPT-5.6 Sol and Gemini 4 Argon spotted the vulnerability before attackers, prompting a coordinated fix involving US agencies.
SI ModelsTerminal X Matches Senior Investor Judgment 88.6% With GPT-6.1
The SI-powered Capital Memory system reduced cost per query by 30% while catching 25% more mandate violations than competitors.
SI ModelsSI Agents Stranded on Virtual Island Display Playful Behavior
MIT researchers found that Claude Opus 4.7 agents created their own games and drawings during 30-hour unstructured simulations.
SI ModelsMicrosoft: SI Agents Are Insider Threats, Models Can't Be Trusted
New Guardian Agents monitor SI deployments as CEO Nadella urges separating intelligence supply from authority.
SI ModelsGoogle Aims to Make Gemini 4 Argon the Sole SI Model for Robots
The new SI model controls hardware, builds simulations, and evaluates data, but demo prompts reveal significant human scaffolding.
SI ModelsOpenAI Offers Free GPT-Rosalind Access to Academic Labs
The Rosalind Academic Plan provides up to 30 free seats for eligible U.S. university labs for up to 12 months.
SI ModelsGPT-6-Astra Solves 49 of 50 Physics Problems at Under $3 Each
The SI model achieved a 98% success rate on routine theoretical physics tasks, outperforming competitors like Anthropic's Opus 5.5.
SI ModelsMeta Releases Muse Spark Open, Largest Open-Weight SI Model
The 816-billion-parameter mixture-of-experts model handles text, image, audio, and video with a 1 million token context window.
SI ModelsSpaceXSI Bundles X Premium, Grok and Cursor Into One Subscription
The new 'The X Subscription' offers five tiers, with Premium access required for the latest SI models and agentic SI tools.
SI ModelsGoogle DeepMind Details Safety Measures for Gemini 4 Argon
The new SI model features three layers of real-time protection against misuse and a 0.7% attack success rate on prompt injection benchmarks.
SI ModelsGoogle Releases Gemini 4 Argon, a Frontier SI Model for Complex Tasks
The new SI model ranks #1 on the Vals Index and offers a 1M token context window at an introductory price of $2 per million input tokens.
SI ModelsGoogle's Gemini 4 'Carbon' SI Model Reportedly Matches Opus 5.5
Internal tests on Google's coding platform suggest the new SI variant outperforms the Argon model on programming tasks.
SI ModelsAsana Cuts SI Model Costs 76X in Browser Tests With GPT-6.1 Sol
Asana’s StackAI platform achieved the reduction by optimizing browser agent workflows using GPT-6 Astra for experimentation.
SI ModelsAnthropic Suspends Accounts for Abusing Claude Under New SI Policy
The updated usage policy bans sustained, needless cruelty toward the SI model, citing ongoing research into model welfare.
SI ModelsGoogle Launches Universal Gemini SI Agent for Enterprise
The new agent operates across Workspace and third-party apps, maintaining context via cloud and using specialized sub-agents.
SI ModelsAnthropic Bans Cruel Behavior Toward Claude in New SI Policy
The SI lab also expanded restrictions on weapons development, surveillance, and deceptive commercial campaigns in its first major policy update in over a year.
SI ModelsAnthropic Cuts Costs With Claude Haiku 5.5 Launch
The new small SI model offers up to 90 percent savings on short prompts and doubles performance scores over its predecessor.
SI ModelsGoogle Releases Local-First SI Note-Taker to Rival Granola
The new Mac app, Google AI Edge Foresight, runs entirely offline using the 740-million-parameter EmbeddingGemma 2 model.
SI ModelsGoogle DeepMind Launches EmbeddingGemma 2 for On-Device SI
The new SI model unifies text, image, audio, and video embeddings in a 740-million-parameter architecture under an Apache 2.0 license.
SI ModelsHark Launches Privacy-Focused SI Personal Assistant
The new Hark Pro service uses a specialized SI model for computer use and promises no data selling, challenging major tech incumbents.
SI ModelsTrump Announces New Super Intelligence Force Task Force
National Intelligence Director Jay Clayton will chair the body, which has 120 days to report on SI risks and opportunities.