Meta's Superintelligence Lab released Muse Wonder and Muse Wonder Full, its first frontier-class models, claiming benchmark scores comparable to Fable 5.1 and GPT-6 Astra, including stronger results on coding tasks. Wonder will be priced at $4 per million input tokens and $20 per million output tokens via Meta's Model API, while Wonder Full is reserved for Muse Code subscribers and won't be offered through the API; both will power the Muse agent.
Meta has released two new AI models, Muse Wonder and Muse Wonder Full, trained at its gigawatt-scale Prometheus data center. The company says the models build on earlier work shown with Muse Spark and post state-of-the-art results on benchmarks covering agents, memory, and coding, including domain-specific tests in law, finance and health such as Harvey's Legal Agent Benchmark.
Samsung is facing a broader memory chip shortage affecting device shipments industry-wide, but strong demand for its Galaxy Z Fold 8 foldable phone has helped the company maintain its position as a top smartphone shipper. The strength of this product lineup is credited with cushioning the company against the wider supply constraints.
A Russian repair specialist known as VIK-off examined an RTX 4090 bought as a bare PCB for about 30,000 rubles ($350) and discovered it contained a fake GPU die soldered on via BGA along with counterfeit memory chips mimicking GDDR6X markings. Closer inspection revealed the die's contact area didn't match a real processor and the solder mask degraded during cleaning, while the memory chips were actually GDDR6 rather than GDDR6X, with VIK-off suggesting the board may actually be a repurposed RTX 3090 Ti PCB.
A developer-focused article walks through arena allocators as an alternative to standard malloc/free memory management in C. It explains how typical manual allocation leads to problems like forgotten frees, memory leaks, and slow performance from many small allocations, before introducing arena allocators as a solution and detailing their implementation.
Apple has raised prices across its MacBook Pro lineup in 2026, affecting even the entry-level configuration compared to 2021 models. The increase follows Apple's broader price hikes announced this year across multiple product lines, departing from the company's past pattern of keeping starting prices relatively stable between generations.
Evan Ovadia has published details of a new borrow checker for the Valen programming language, which he describes as more flexible than existing systems. The design uses a single kind of borrow reference rather than the two typically found in other borrow checkers, and includes a feature that lets the compiler track where a reference points. Some syntactic features, such as auto-deref and direct struct indexing, are planned for a November release.
RAM temporarily stores data a processor needs quick access to, using capacitors that lose their charge once power is cut, unlike ROM which retains data permanently. TrendForce reports that DRAM contract prices jumped 90 to 95 percent quarter-over-quarter in early 2025, driven largely by AI data center buildouts such as NVIDIA's proposed $100 billion investment with OpenAI for 10 gigawatts of capacity. Micron also shut down its consumer-focused Crucial brand in February to prioritize higher-margin AI data center business.
A shortage of computer RAM, informally dubbed 'RAMageddon,' is being driven in part by AI data centers consuming large quantities of memory chips. The shortage is pushing up prices for consumers and businesses that rely on RAM for computers and other devices.
Cloudflare has acquired Deno, the JavaScript runtime and development environment, and plans to wind it down over the next year. The author, a longtime user who built a business on Deno, Deno Deploy, Deno Sandboxes and Fresh, says migration to Node or Bun is straightforward since Deno never created strong lock-in.
Motherboard maker Gigabyte indicates Intel plans to release new CPUs next year using the older LGA 1700 socket, which supports both DDR4 and DDR5 memory. Intel VP Robert Hallock said Raptor Lake chips will remain available for 'years to come' and that LGA 1700 remains viable because demand for DDR4 persists. A 32GB DDR4 kit costs around $210 versus roughly $480 for a comparable DDR5 kit.
AMD's CEO Lisa Su is visiting South Korea to negotiate additional HBM memory supply for AI hardware, following a recent memorandum of understanding with Samsung. Samsung is also interested in broadening its collaboration with AMD to include foundry manufacturing and other semiconductor technologies. The discussions highlight ongoing efforts to secure long-term supply chains amid rising AI demand.
GoKawiil