OpenAI Gives 100,000 Researchers a Free Year of ChatGPT‑Powered Scientific Workflow

OpenAI Free Academic Program On July 29, OpenAI launched a one‑year free‑access program for 100,000 academic researchers, starting with 10,000 slots for institutions such as École Polytechnique and Princeton. Selected users receive the flagship GPT‑5.6 Sol Pro model and a full suite of research tools covering literature review, code generation, data analysis, paper writing, and grant applications. The plan includes expanded context windows, specialized life‑science skills, and integration with third‑party research tools, while guaranteeing that user data will not be used to train the models.

The article notes that AI‑assisted research is now mainstream: OpenAI reports over one million weekly users conducting scientific work with ChatGPT, and the number of arXiv papers acknowledging ChatGPT in mathematics continues to rise. Long‑term, OpenAI aims to embed researchers in its ecosystem, creating migration costs that may keep users after the free year, a strategy contrasted with Anthropic’s more limited support.

GPT‑5.6 Pricing Overhaul OpenAI announced a major price reduction for its Luna and Terra models, cutting overall inference costs by 80 % for Luna (the high‑speed, low‑cost tier) and 20 % for Terra (the balanced tier). The flagship Sol model’s price remains unchanged. The cuts are attributed to internal AI‑driven optimizations: autonomous analysis of traffic patterns, load‑balancing improvements, GPU kernel rewrites, and cache‑policy refinements, all validated through extensive safety checks.

Benchmark results show Luna now outperforms Anthropic’s high‑end models in cost‑per‑task metrics and runs faster in code‑generation tests, while a new “fast mode” adds latency reductions without sacrificing model quality.

TRAE Work Enterprise AI Platform TRAE reported 10 million registered users and over 5.6 trillion daily tokens processed. The platform launched TRAE Work to address AI adoption across non‑engineering roles, providing standardized development rules, a shared knowledge base, and visual performance metrics. It supports product, operations, and design teams by generating documents, prototypes, data reports, and brand‑compliant designs, all while enforcing security through encrypted transmission, tenant isolation, and fine‑grained permissions.

Qualcomm Personal‑AI Vision Qualcomm’s IDC‑backed white paper shows declining shipments of smartphones, PCs, and tablets but rising AI‑feature penetration. To resolve value‑misalignment, Qualcomm proposes a distributed edge‑cloud architecture that dynamically routes lightweight tasks to the device, medium tasks to edge nodes, and heavy tasks to the cloud, leveraging heterogeneous hardware (CPU, NPU, GPU) and on‑device knowledge graphs to preserve privacy and reduce latency.

GCC AI‑Code Contribution Policy The GNU Compiler Collection announced a policy restricting AI‑generated core code contributions due to copyright and accountability concerns. Non‑critical code may be accepted if the AI contribution is clearly marked; test‑case code is exempt. The policy reflects broader community differences, with Linux allowing AI assistance under strict responsibility, while projects like Zig ban AI contributions outright.

Claude Code Prompt‑Tuning Insights Anthropic’s Claude Code founder Boris Cherny explained that the recent 80 % reduction of system prompts in Opus 4.8 was a deliberate cleanup of redundant rules that constrained model autonomy. New prompts in Opus 5 focus on bounding task scope and limiting unnecessary error‑explanations, illustrating the industry lesson that over‑engineering prompt libraries can hinder model evolution.

Anthropic’s Product Management Shift Diane Penn, Anthropic’s first technical PM, described how the team treats evaluation suites as the new product requirement documents. By quantifying user pain points into standardized test sets, they drive iterative model improvements and maintain a rapid experiment‑first culture while balancing safety and security.

Li Fei‑Fei’s Robot‑School Initiative World Labs acquired simulation startup SceniX to build a digital training environment for robots. The combined platform merges Marble’s 3‑D world generation with high‑fidelity physics simulation, enabling large‑scale, low‑cost robot training that bridges the gap between virtual testing and real‑world deployment.

Overall, the roundup highlights a week of significant AI cost reductions, ecosystem‑building initiatives, and policy developments shaping the future of large‑model deployment and AI‑augmented workflows across research, industry, and open‑source communities. (Sources: 新智元, CSDN, 量子位, 网易科技, 硅星人Pro, PaperWeekly)