In partnership with
In Today’s Issue:
💰 ChatGPT Mobile hits a record $3B in revenue
💻 GPT-5.2-Codex launches as a long-horizon powerhouse for autonomous engineering
📄 Mistral OCR 3 outperforms previous models on messy scans
🛡️ Anthropic hardens Claude’s safety layer
✨ And more AI goodness…
Dear Readers,
If AI coding is going to matter, it has to finish the messy work, not just autocomplete. GPT-5.2-Codex is OpenAI’s swing at exactly that: longer-horizon agentic changes, improved reliability in Windows environments, and top-tier results on real-world benchmarks like SWE-Bench Pro and Terminal-Bench 2.0.
In the same "make it real" vibe, Mistral OCR 3 targets the "PDF prison" with structured markdown/table reconstruction and aggressive pricing ($2 per 1,000 pages). Anthropic is also fortifying Claude’s emotional-support guardrails—especially around self-harm—and tying them to broader auditing work like the Petri benchmark.
Meanwhile, the market signal is loud: ChatGPT’s mobile app just hit $3B in consumer spending in 31 months. And at the macro level, OpenAI’s new DOE collaboration around the Genesis Mission makes it clear the U.S. is treating frontier AI as national infrastructure.
All the best,
Kim Isenberg
GPT-5.2-Codex Pushes Real-World Engineering Forward
GPT-5.2-Codex is OpenAI’s newest “agentic” coding model, tuned for long-horizon work (context compaction), big repository changes (refactors/migrations), better Windows reliability, and stronger screenshot/diagram understanding. It hits state-of-the-art results on SWE-Bench Pro and Terminal-Bench 2.0.
It’s rolling out now across Codex for paid ChatGPT users (API access planned “in the coming weeks”). You can start via the Codex CLI (npm i -g @openai/codex), with sandboxing that defaults to no network access and per-project, user-controlled allow/deny lists.
Mistral sets new ORC-standard with Mistral OCR 3
Mistral OCR 3 is Mistral’s latest document-understanding model, claiming a 74% overall win rate vs. OCR 2 on tough inputs like forms, low-quality scans, complex tables, and handwriting—while outputting clean text plus structure (markdown with HTML-style table reconstruction).
It’s available via API (mistral-ocr-2512) and in Mistral AI Studio’s Document AI Playground, priced at $2 per 1,000 pages (or $1 per 1,000 pages with the Batch API discount).
Anthropic Hardens Claude’s User Wellbeing
Anthropic lays out how Claude handles sensitive emotional-support chats—especially regarding suicide and self-harm—and backs it with product and training safeguards. This includes a dedicated classifier that triggers a crisis banner with country-specific resources via ThroughLine’s network (170+ countries), plus new expert input via the International Association for Suicide Prevention.
On evals (run without the system prompt), Anthropic reports very high “appropriate response” rates in clear-risk single-turn cases—98.6% (Opus 4.5), 98.7% (Sonnet 4.5), 99.3% (Haiku 4.5)—and improved multi-turn performance. They are also pushing harder against “sycophancy” and delusion with audits and the open-source Petri benchmark, while reaffirming Claude.ai’s 18+ requirement and underage-detection work.
Sam Altman: How OpenAI Wins, AI Buildout Logic, IPO in 2026?
ChatGPT Becomes a Mobile Giant:
A big step in the right direction
The Takeaway
👉 ChatGPT’s mobile app has now hit $3B in worldwide consumer spending (iOS + Android) since launching in May 2023, based on Appfigures estimates.
👉 2025 did the heavy lifting: about $2.48B spent in 2025 vs $487M in 2024 and $42.9M in 2023 (a sharp acceleration in paid adoption).
👉 ChatGPT reached $3B in 31 months, faster than TikTok (58 months) and major streamers like Disney+ (42) and HBO Max (46).
👉 Spending is driven largely by subscriptions (e.g., $20/month Plus and $200/month Pro), with additional monetization paths discussed (like ads) and a newly launched “app store”-style directory that could open future revenue streams
ChatGPT just crossed a wild consumer milestone: $3 billion in lifetime mobile spending in only 31 months. That’s not “AI hype” money; it’s people pulling out their phones and paying for a tool that’s becoming as habitual as maps or messaging.
Fresh app-economy estimates suggest most of that cash landed this year: about $2.48B in 2025 alone, up 408% from roughly $487M in 2024 (after a $42.9M start in 2023). In speed, ChatGPT beat even mobile giants: TikTok needed 58 months to hit $3B, while Disney+ and HBO Max took 42 and 46 months.
What’s driving it? Subscriptions. Users aren’t just “trying” AI—they’re budgeting for it, paying monthly to get faster models, higher limits, and pro features. Next comes distribution: OpenAI is rolling out an in-ChatGPT app directory where developers can publish chat-native apps and, over time, unlock more monetization paths. If that flywheel works, “ChatGPT on your phone” starts looking less like a single product and more like the front door to a new software layer.
Why it matters: This is proof that consumers will pay real money for AI when it reliably saves time and friction. And it hints that the next platform shift may be “apps inside AI,” not just AI inside apps
Sources:
🔗 https://techcrunch.com/2025/12/18/chatgpts-mobile-app-hits-new-milestone-of-3b-in-consumer-spending/
Turn AI Into Extra Income
You don’t need to be a coder to make AI work for you. Subscribe to Mindstream and get 200+ proven ideas showing how real people are using ChatGPT, Midjourney, and other tools to earn on the side.
From small wins to full-on ventures, this guide helps you turn AI skills into real results, without the overwhelm.
Even closer cooperation between the US government and AI companies: The Genesis Project
When frontier AI meets national-lab horsepower, science can iterate more like software: propose, test, learn, repeat. OpenAI just signed a memorandum of understanding (a formal “let’s build together” framework) with the U.S. Department of Energy to collaborate on AI and advanced computing in support of DOE initiatives, including the Genesis Mission—an effort to use AI to accelerate discovery science, strengthen national security, and drive energy innovation.
This isn’t just a shiny demo. Over the past year, OpenAI and DOE national labs ran a “1,000 Scientist AI Jam Session” across nine labs, stress-testing frontier models on real research tasks and turning the feedback into better systems. OpenAI has also deployed advanced reasoning models on Los Alamos’ Venado supercomputer as a shared resource across the NNSA lab network (Los Alamos, Lawrence Livermore, Sandia). On bioscience, OpenAI and Los Alamos are developing more realistic safety evaluations for multimodal models in laboratory settings, moving beyond text-only tests.
National labs are where AI gets calibrated against reality, not benchmarks. Faster, safer feedback loops mean more breakthroughs move from “interesting” to “validated” sooner.
It is clear that the US is deploying all its resources to maintain its dominance, particularly against China. To this end, it is supporting major national players like OpenAI with all necessary resources.
Sources:
🔗 https://openai.com/index/us-department-of-energy-collaboration/
🔗 https://genesis.energy.gov
