AI Cookbook
← Back to blog

AI Weekly — GPT-5.5, Codex Labs, Claude Opus 4.7, and Project Glasswing

OpenAI and Anthropic both shipped heavy updates this week. GPT-5.5, Codex hitting 4M+ weekly devs, Opus 4.7 generally available, and the new Mythos Preview focused on offensive security — everything in a 5-minute read.

1. GPT-5.5 — what changed

OpenAI launched GPT-5.5 with improvements on three fronts: coding (more accurate on long tasks), computer use (autonomous agent capabilities), and deep research (more consistent reasoning chains in multi-step research).

The model is already rolling out to Plus, Pro, Business, and Enterprise on ChatGPT and Codex. For API users, pricing stays close to the prior 5.0 — Sam Altman declared they want broad access, not premium pricing.

2. Codex Labs and the dev adoption curve

In early April, more than 3 million devs were using Codex weekly. Two weeks later, more than 4 million. OpenAI launched Codex Labs and closed partnerships with Global Systems Integrators (GSIs) to distribute Codex to enterprise.

Honest read: the coding-assistant adoption curve is steepening faster than the general chat curve. For anyone creating tech content, this is a niche graduating from early adopter to corporate production.

3. Anthropic — Claude Opus 4.7 generally available

Anthropic released Claude Opus 4.7 in general availability. Visible gains: better software engineering, multi-turn coding tasks without losing the thread, and improved vision with higher-resolution image support.

Pricing matches Opus 4.6: $5 per million input tokens, $25 per million output. Good budget stability — no upgrade tax this cycle.

4. Mythos Preview + Project Glasswing

On April 7, Anthropic introduced Claude Mythos Preview: a general-purpose model particularly strong on computer security tasks — vulnerability analysis, defensive code reasoning, and so on.

In parallel, they launched Project Glasswing — a coordinated effort to use Mythos Preview to harden critical real-world software. It is not a separate consumer model; it is a public signal that AI-security has become an explicit institutional priority for Anthropic.

Takeaways

  • Top models (GPT-5.5, Opus 4.7) reward anyone who revisits old prompts
  • Coding is moving from experiment to corporate production
  • Security has its dedicated model for the first time
  • Stable pricing on Opus = time to optimize prompts, not cut costs

Sources

  • OpenAI — Introducing GPT-5.5
  • Anthropic — Introducing Claude Opus 4.7
  • Anthropic — Claude Mythos Preview & Project Glasswing
  • TechCrunch — OpenAI releases GPT-5.5