It’s Topical
← The AI Builder Roundup archive
The AI Builder Roundup

Build apps with your ChatGPT plan's API credits

Hi everyone, your ChatGPT subscription may now cover some of the API calls behind your next app. We're breaking down who gets credits, plus a coding model trained to finish tasks with less reasoning and fewer output tokens.

In today’s edition

  • ChatGPT subscriptions now include eligible API credits.
  • Particle launches a more token-efficient coding model.

Your ChatGPT plan can now help fund your app

Eligible ChatGPT Pro and Business subscribers can now claim recurring OpenAI API credits. That gives you an API budget for building apps and agents alongside your existing subscription.

  • Pro plans receive $100, $200, or $500 per month, depending on the tier.
  • Business plans receive $20 per paid Standard seat or $100 per paid Premium seat, pooled up to $500 per workspace per month.
  • Your first grant is prorated and expires at the end of your subscription billing cycle. Business workspaces paying annually upfront receive the year's grant in one issuance.

What it means Claim the benefit in ChatGPT billing and link an API organization you own before buying more API credits. It doesn't cover Codex usage through API keys.

Source: OpenAIDevs

Particle-1 targets the extra tokens in coding tasks

Particle launched a coding model post-trained on DeepSeek V4.1 Flash to solve tasks with fewer output tokens. You can use it through the Particle API or OpenRouter.

  • Across four coding benchmarks, Particle reports 28.4%–52.2% fewer output tokens and 34.4%–51.6% lower average cost per task than the base model, with pass rates within its confidence intervals.
  • In one successful build example, it inspected the Makefile before compiling instead of repeating installation and verification. That reduced the run from 18 to 12 model calls.
  • OpenRouter lists pricing at $0.25 per million input tokens and $1 per million output tokens, with a 1.05M-token context window.

What it means Try it on a repeatable coding task in your existing agent workflow. Compare completed work and total task cost against DeepSeek V4.1 Flash.

Source: particle.ai

Quick hits

  • Google announced Gemini 4 Carbon with claimed gains in cybersecurity and sustained agent work, initially restricting access to selected trusted cyber defenders through Fairwind.
  • Replit released ViBench 2.0 and found nearly three-quarters of top models' failed builds broke only where features interacted; eight products and the harness are public.
  • GitHub Copilot's app previews Projects for coordinating agents across repositories, plus computer use on macOS and Windows that asks for approval before controlling an app.
  • GitHub Agentic Workflows reached general availability for defining repository automations in Markdown that compiles to Actions YAML; the default Copilot CLI engine requires a Copilot license.
  • Vercel CI entered beta for Turborepo projects, running only tasks affected by code changes; Vercel reports 2–6× shorter push-to-pass waits in its internal repositories.

GitHub repo of the day

Claude-Mem: carry project knowledge between sessions

  • What it does: The maintainers describe a memory system that records tool activity and summaries so Claude Code can recall past fixes and decisions across sessions.
  • How to start: With Node.js 20+ and Claude Code plugin support, run npx claude-mem install, sign in, choose a memory provider, then restart Claude Code. The hosted observer has a 30-day trial and falls back to your Anthropic plan afterward.

AI workflow of the day

How to find what's using your Grok Bot allowance

Shared by Robin Delta (@heyrobinai) · View original post ↗

If your Grok Bot allowance runs out while you're away, check scheduled routines before paying for more usage. This method helps you find repeated background work and decide what to pause or run less often.

  1. Open Settings, then Usage. Check when your allowance resets, which bots and routines used the most, and which day usage jumped.
  2. Alternatively, ask Grok Bot to read the breakdown using the prompt below. Don't create a scheduled routine just to monitor usage, since those checks use allowance too.
  3. Run routines less often when their results don't need to be fresh. Restrict schedules to useful hours, or replace a schedule with a Slack, email, or GitHub trigger when appropriate.
  4. Merge routines that gather the same information. Pause work you don't need instead of deleting it, so you can bring it back later.
Original prompt
More from The AI Builder Roundup →