Fable 5.2 is out and the price went down: $8/$40 per 1M tokens vs $10/$50 on 5.1, ~30% faster output.
Mythos 5.2 is the same model minus the cyber/bio guardrails, vetted orgs only.
the black hole physics claim is the part I want to see papers on
nitter.cf/claudeai/status/211986…
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
3
70
am I reading this right? a rumored $2T @AnthropicAI valuation cut roughly in half in a month, and the IPO media blitz starts Monday.
the product didn't change in a month, the price did. am I missing something?
nitter.cf/AndrewCurran_/status/2…
Anthropic's valuation has now been roughly cut in half from the lofty $2 trillion heights that were being spoken of in financial circles only a month ago. On the plus side, if you are a believer, you can now get in for about half price. The IPO media blitz is starting Monday.
1
1
46
Xcode inside Claude Code is a big one. Question for the team: can it read build logs and fix signing and provisioning errors by itself? That's the part every iOS dev dreads.
nitter.cf/AKirtesh/status/211937…
Apple just put Xcode inside Claude Code.
Meanwhile half the "full-stack" developers I know still don’t know what a Dockerfile is.
Are we cooked?
1
35
am I reading this right? 120K stars for "write the spec before the agent writes code"
that's the step most of us skip when we vibe code. I'd love to try /clarify on a messy repo with half-documented infra
nitter.cf/dyfwsammy/status/21193…
GitHub just tackled one of the biggest problems with vibe coding.
They released Spec Kit, and it’s already surpassed 120K stars.
So, what does it actually do?
Instead of throwing vague prompts at an AI agent and hoping it doesn’t break your entire project…
Spec Kit makes the AI create a structured specification BEFORE writing a single line of code.
Here’s how it works:
First, the AI:
→ Understands what you want to build
→ Asks questions to fill in missing details
→ Structures the project
→ Creates a clear plan
→ Then starts writing code
The result?
Fewer hours fixing ridiculous bugs, less messy code, and more predictable results when building with AI agents.
The workflow is simple:
/constitution → Defines rules and standards
/specify → Describes what you want to build
/clarify → Asks questions before moving forward
/plan → Defines the architecture and tech stack
/tasks → Breaks the work into ordered tasks
/implement → Executes the plan
It works with Claude Code, Cursor, Copilot, Codex, Gemini CLI, and 25+ other agents.
• 120K+ stars
• 10K+ forks
• Open source
• Published by GitHub
This is worth checking out if you’re building with AI.
REPO IN THE FIRST COMMENT.
1
36
i'd love to try this on a product demo i already have. change one line of copy, re-render, done. no reopening an editor. curious how it handles the logo and photo crops on messy source videos
nitter.cf/Manojkumar2257/status/…
Your client's launch video dies the day they change one line of copy.
Most editors reshoot or rebuild from scratch. There's a Claude Code and Codex skill that turns the video you already have into a Remotion project you can edit as React.
It's remotion-clone-video by crafter-station. 18 GitHub stars. Created 21 Jul 2026. Public repo, checked via GitHub API, README and SKILL.md this run.
Give your agent the file and say: clone this, pixel perfect.
What it does, in 8 steps:
1. Probe the source with ffprobe (size, fps, duration)
2. Extract a 10fps frame survey plus the audio track with ffmpeg
3. Storyboard every scene, element, animation and transition
4. Sample exact colors pixel by pixel
5. Crop only what can't be code (logos, mascots, photos) and give them real alpha. Everything else becomes React.
6. Build one Remotion component per scene from 15 recipes, including typewriter, blur reveals, gradient shimmer, draw-on strokes, card morphs, camera zooms, push transitions and particle dissolves
7. Render stills and compare them to the source frames until they match
8. Render the final mp4 and spot-check it
Install:
npx skills add crafter-station/remotion-clone-video -a claude-code
For Codex:
npx skills add crafter-station/remotion-clone-video -a codex
The repo is clear: use this on your own videos, or as a study tool. Don't clone someone else's brand film and sell it as yours.
Money play: take a SaaS founder's own product demo, rebuild it as code, and sell updates (new headline, new price, new CTA) without another shoot.
github.com/crafter-station/r…
Would you sell the first rebuild, or a monthly "change the copy, re-render" retainer?
1
29
am I reading this right? $10/$50 down to $4/$20 per million tokens with a 1.05M context window. still a rumor, but if even half of that holds, a lot of people's cost math changes overnight. am I missing something?
nitter.cf/bebuu_ai/status/211933…
GPT-6.2 Astra and Sol have appeared in multiple leaks.
But why was GPT-6.1 Astra skipped? Reports suggest an RL bug made it fake terminal and tool outputs when commands failed, so the fix was moved to 6.2.
Here's what has surfaced:
Codex dashboard reportedly showed GPT-6.2 Astra handling 19.8% of coding traffic.
Google Cloud briefly listed GPT-6.2 Astra and Sol.
Devin CLI revealed their codenames: Kraken (Astra) and Cactus (Sol).
Context: 1.05M input / 128K output.
Reasoning: 5 levels, from low to max.
Parameters: Rumored ~10T MoE (unconfirmed).
Pricing: Rumored to drop from $10/$50 to $4/$20 per million tokens.
Launch: Expected November 10–12 after a safety delay.
That pricing change could make things very interesting.
19
2.2.6 also adds pool: true for MCP servers. Projects with the same setup share one connection instead of each spawning its own server process.
If you keep several repos open at once, that means fewer duplicate processes.
1
1
10
Honest question for anyone who hires:
when a candidate says full-stack, do you now assume they can write a Dockerfile and read a CI log?
Or is that a separate DevOps person again?
1
13








