
Tech • IA • Crypto
OpenAI’s preview-only GPT‑5.6 signals a more powerful yet tightly controlled AI era, as governments, rival models, sovereign data tools, autonomous weapons, and new infrastructure bets reshape the tech landscape.
OpenAI introduced GPT‑5.6 in a limited preview for select partners, with no firm public release date. The model is not yet accessible in ChatGPT, reflecting caution after recent industry controversies around powerful systems. The rollout suggests uncertainty about broad deployment despite claimed performance gains.
The system is split into Sol (high-end), Terra (balanced), and Luna (fast, lightweight). Sol Ultra reportedly uses hidden “sub-agents” to handle complex reasoning tasks in parallel, hinting at more autonomous workflows. This structure closely mirrors competing model families from Anthropic, underscoring converging design trends.
OpenAI positions GPT‑5.6 Sol near top-tier competitors on benchmarks while using up to three times fewer tokens, suggesting improved efficiency. Pricing remains similar to earlier models, with $5 input / $30 output per million tokens for Sol. Whether efficiency reflects optimization or lower raw capability remains unclear.
Development involved close coordination with the U.S. government, including input into potential regulatory frameworks. The model is classified as high risk in cyber and bio domains but not “critical.” OpenAI reports 700,000+ hours of testing and claims a 0% jailbreak success rate, though such guarantees remain debated.
A “prompt caching” system allows reuse of repeated queries, cutting costs by up to 90% on reads. Future deployment may leverage Cerebras wafer-scale chips, aiming for faster inference and reduced latency. This points to a strategy combining software efficiency with hardware innovation.
France introduced carte.gouv.fr, a state-backed geospatial platform offering 1,100+ data layers, from demographics to infrastructure and environment. Built by IGN, it replaces older systems and emphasizes open data and collaboration, including a crowd-sourced street-level imagery system called Panoramax.
The platform enables developers and businesses to combine datasets for applications in urban planning, real estate, agriculture, and energy. Its open architecture positions it as a sovereign alternative to Google Maps, with deeper analytical capabilities rather than simple navigation.
In Ukraine, AI-enabled drone systems are increasingly used for long-range strikes, interception, and coordination in swarms. Models embedded onboard allow navigation without GPS and real-time decision-making. Systems like Hornet and Ruta demonstrate growing autonomy, with humans reduced to initial targeting decisions.
These systems improve efficiency and survivability but raise concerns about escalation and loss of human control. The spread of autonomous swarm capabilities could trigger broader geopolitical competition, with major powers racing to match or counter such technologies.
Tesla filed a detailed trademark for Megapod, described as modular AI data centers integrated with its Supercharger energy network. The concept suggests deploying compute infrastructure at scale using existing power capacity, potentially creating a decentralized AI backbone.
With access to gigawatts of distributed energy, Tesla could position itself as a provider of both power and AI infrastructure, competing with traditional cloud providers. The move aligns with broader ambitions to control key layers of the AI supply chain.
Across AI models, infrastructure, and geopolitics, innovation is accelerating while control, sovereignty, and risk management increasingly shape what progress actually reaches the public.