Tech • AI • Robotics • Game
Google's Project Suncatcher Is Sending AI to Space!?, Sonnet 5.5, $500 ChatGPT Plan & More! AI News
9.4/10TL;DR
Anthropic, OpenAI, and Google are all preparing major AI moves, from a rumored Claude Sonnet 5.5 launch and a possible $500 ChatGPT Pro Max tier to internal math breakthroughs and orbital AI computing tests.
KEY POINTS
Claude Sonnet 5.5 leak points to aggressive pricing
Anthropic is reportedly testing Claude Sonnet 5.5 with select partners and may pair it with unusually low pricing for a frontier model. The leaked structure suggests $2 per million input tokens, $10 per million output tokens, and $0.20 per million cached-read tokens. The model is also said to support a 1 million-token context window and up to 128,000 output tokens, which would make it one of the most expansive mainstream offerings if released as described.
Early previews suggest stronger creative and coding output
Previewed results indicate Sonnet 5.5 is being positioned as a serious rival to GPT-6 Sol, especially in structured visual generation and code-heavy tasks. Examples included detailed SVG renderings of game controllers with lighting and texture effects, along with a functional Minecraft-like sandbox featuring caves, crafting, fall damage, gravel physics, mobs, and item drops. The outputs were not flawless, but they suggested stronger execution and polish than many lightweight coding demos.
Reasoning defaults may change how Claude is used
The leaked setup indicates that reasoning in Sonnet 5.5 is adaptive by default. Fully disabling reasoning may only be available at lower effort levels rather than at the highest tiers. That design would signal a push toward higher baseline performance without forcing users to manually tune every request.
OpenAI may introduce ChatGPT Pro Max at $500 a month
OpenAI is reportedly preparing a new subscription tier called ChatGPT Pro Max priced at $500 per month. References found in development point to features framed around faster work and Codex, with speculation that upgraded infrastructure could deliver faster performance and possibly higher usage caps. The move follows earlier reports that OpenAI had explored even more expensive premium plans.
Premium tier may target long-running and coding-heavy tasks
The rumored Pro Max offer appears aimed at power users who need more sustained compute, especially for coding and extended workflows. OpenAI has also been developing systems for longer-running tasks, so the new tier could become the first subscription built around that capability. More details may emerge around the company’s upcoming developer announcements.
OpenAI says an internal model is solving major math problems
OpenAI has disclosed that an internal model that began training on August 28 has reportedly resolved more than 100 long-standing open mathematical problems across several areas. The same effort is also tied to a claimed advance on the Navier-Stokes problem, one of the best-known unsolved questions in mathematics and physics. The scale of the claim has prompted unusual scrutiny.
Independent mathematicians are being brought in
To assess the significance of these results, OpenAI is working with an independent advisory group that includes Timothy Gowers, Martin Hairer, Avi Wigderson, and Edward Witten. Their role is to evaluate the mathematical importance of the AI-generated work and help determine how it should be communicated to the research community. The step reflects the difficulty of verifying frontier AI claims in advanced mathematics.
Google is testing AI chips for use in space
Google has announced Project Suncatcher, a moonshot effort to explore whether large-scale AI compute could eventually be hosted in orbit. The company is preparing a prototype satellite carrying TPUs into low Earth orbit, where near-constant sunlight could provide far more solar energy than comparable terrestrial systems. Google’s long-term vision is a network of AI satellites linked together to process workloads in space.
Project Suncatcher faces launch, radiation and cooling hurdles
Google says hardware for orbital AI must survive launch forces of roughly 50 to 100 g, radiation exposure, and the absence of air cooling. The company has already tested Trillium TPUs under proton radiation while they were running AI workloads and says the chips tolerated more exposure than expected over a five-year mission. Because heat cannot be removed like it is in a normal data center, Google is experimenting with heat pipes and radiators.
Laser-linked TPU clusters are planned for 2027
Future satellites could carry dozens of TPUs each and communicate using high-bandwidth laser links. Google describes the alignment challenge as similar to hitting a coin-sized target from miles away while both objects are moving. A first orbital TPU test is planned before two larger satellite experiments in 2027 focused on interconnection technology.
Anthropic adds a fallback for Claude rate limits
Anthropic is also offering a practical workaround for users who hit Claude Opus 5.5 rate limits. A low-priority mode can continue processing requests when spare capacity is available, though at slower speeds. Separate testing has also suggested that low reasoning effort can perform close to maximum settings on many everyday tasks while costing about 12 times less, making it a cheaper default for routine work.
CONCLUSION
The latest AI race is widening on several fronts at once: cheaper frontier models, premium subscriptions, scientific discovery claims, and even orbital infrastructure. If the reported capabilities and pricing hold, the next competitive phase in AI will be defined as much by access and compute strategy as by model quality.
Ask a question