In partnership with

Welcome back!

The big launch is finally here: GPT‑5 has officially arrived, and it’s more than just an incremental upgrade. It replaces all previous ChatGPT models, powered by a new router system that selects the best mode for the current task. It ranks #1 on LMArena, outperforms rivals in coding, reduces hallucinations, and shows early signs of being more reliable in many real-world scenarios. Let’s unpack…

Today’s Summary:

• 🔥 OpenAI releases GPT‑5

• 💻 GPT‑5 sets new coding standard

• 📈 Where GPT‑5 actually changes the game

• 🌍 AlphaEarth maps planet via AI

• 🏛️ ChatGPT reaches U.S. government

• ⚙️ Google debuts Gemini CLI GitHub Actions

• 🛠️ 2 new tools

TOP STORY

GPT-5 becomes ChatGPT universal engine

The Summary: OpenAI has released GPT-5, its most capable AI model to date, to all ChatGPT users, with Pro and Plus subscribers getting expanded access and GPT-5 Pro for extended reasoning. The new system replaces previous models and can switch automatically between fast and deep-thinking. GPT-5 aims to deliver faster, more accurate, and more trustworthy responses while reducing hallucinations and over-flattery.

Key details:

• Now the default for all logged-in users, replacing GPT-4o and o3

• No model picker required: GPT-5 automatically switches between fast mode and deeper thinking mode based on task complexity. You can force a mode by prompting, eg “think hard about this”

• Free users get 10 GPT-5 messages per 5 hours and one Thinking message per day before switching to GPT-5 Mini

• Plus users get 80 GPT-5 messages per 3 hours and up to 200 Thinking messages per week

• Old conversations are auto-mapped to the closest mode

• Supports all ChatGPT tools, including web search, data and image analysis, file analysis, and image generation

• New #1 rank in LMArena at 1481 pt, beating Gemini 2.5 Pro at 1460

• Outperforms previous models on math benchmarks such as AIME 2025 (94.6% without tools)

• Four opt-in chat personalities in ChatGPT settings: Cynic, Robot, Listener, and Nerd

• Rolling out globally to Plus, Pro, and Team, with others coming soon

Why it matters: The launch strengthens OpenAI’s lead in general-purpose AI, delivering a model that not only scores high in benchmarks but also adapts to task complexity. Lower error rates and faster reasoning mean more reliability for high-stakes uses, from professional research to everyday assistance. Competitive pricing pressures rivals, further accelerating accessibility of top-tier AI.

FAQ

OPENAI

GPT-5 sets new standard for coding

The Summary:  GPT-5 is OpenAI’s strongest coding model yet, topping benchmarks in software engineering, code editing, and agentic development. It can generate high-quality code, debug large repositories, and run complex workflows reliably. Early testers are calling it “the first model we can actually rely on”.

Key details:

• State-of-the-art coding scores: 74.9% on SWE-bench Verified (a little more than Claude Opus 4.1 at 74.5%), 88% on Aider Polyglot, and preferred over Claude and Gemini in side-by-side dev workflows

• Understands entire codebases: 400K token context, sufficient for entire monorepos or logs in one pass

• Frontend: handles correct spacing, layout, typography, and interaction

• Agentic coding: Handles tool errors, recovers mid-process, and strings multi-step tasks without breaking

• API pricing: GPT‑5 : $1.25 in / $10 out per million tokens, GPT‑5 Mini : $0.25 / $2, GPT‑5 Nano : $0.05 / $0.40

• Alpha feedback from dev tools: Cursor, Vercel, Windsurf, GitHub, Manus all report major boosts in tool reliability, code accuracy, and agent stability

Why it matters: GPT‑5 is a strong coding model which starts behaving like a partner, for planning, decisions, and build, with a clarity of the “how” and the “why” of complex tasks. For engineers it starts to feel like a model ready for production workflows.

FROM OUR PARTNERS

AI Content Creation Without the Burnout

Go From Idea to Video in Minutes—Not Hou

Creating content daily shouldn’t feel like a full-time job.

Syllaby.io helps you generate faceless videos in just minutes—no editing, no filming, and no burnout.

✅ Auto-generate engaging short or long-form scripts
✅ Add captions, voiceovers, B-roll & character-consistent avatars
✅ Schedule and publish to TikTok, YouTube, Reels, and more

Whether you're a solopreneur or agency, Syllaby gives you everything you need to scale your content—fast.

Get 7-days free!

OPENAI

Where GPT‑5 may actually change the game

The Summary: GPT‑5 biggest leap could be in trustworthiness. The model got major upgrades in factual accuracy and strongly reduces errors and deceptive answers. Thus GPT‑5 may become more usable for real-world scenarios like healthcare patient prep, strategic writing, expert-level tutoring, or decision support.

Key details:

• Top score on HealthBench Hard: 46.2% vs 15.8% for GPT-4o. This benchmark tests realistic medical queries against physician criteria

• ~80% fewer hallucinations vs GPT-4o, critical for legal and factual tasks; 6x drop in hallucinations vs o3 on factual prompts

• Deception rate in production cut from 4.8% (o3) to 2.1%, with better recognition of impossible or underspecified tasks

• Maintains accuracy over 400K-token contexts

• New safety system with nuanced answers for sensitive or dual-use queries

• Research and tutoring-ready: GPT‑5 ranks 88.4% on GPQA Diamond PhD-level science questions

Why it matters: Judgment is where trust lives. In health, legal, education, or strategic tasks, users need clarity and no made-up facts. GPT‑5 starts showing signs of that. By being able to say “I can’t” when needed it becomes more viable for regulated domains and high-stakes decisions. It won’t replace professionals but it has now the skill to support them more responsibly.

QUICK NEWS

Quick news

• Google introduces Gemini CLI GitHub Actions for team collaboration

• ChatGPT rolls out to the entire U.S. federal workforce

• AlphaEarth maps the planet with AI precision

TOOLS

🥇 New tools

• Ideogram Character - Character consistency with just one image

• Nitrode - AI game engine to prototype 3D games