OpenAI cuts Astra API prices with GPT-6.2
OpenAI released GPT-6.2 Astra and Sol, with lower standard API token prices for Astra and access through ChatGPT, Codex, and the API.
In today’s edition
- OpenAI releases GPT-6.2 Astra and Sol.
- Devin Bots manage ongoing coding goals.
- Google AI Studio adds Android data sync.
GPT-6.2 Astra costs less through the API
OpenAI has released GPT-6.2 Astra and Sol for ChatGPT, Codex, and the API. Astra’s standard input and output token prices are 20% lower than its predecessor’s.
- For shorter API requests, Astra costs $8 per million input tokens and $40 per million output tokens. Sol costs $2 and $10, respectively.
- OpenAI reports that on Cognition’s FrontierCode 2.0 test at maximum reasoning, Sol came within half a point of Astra at roughly one-fifth the API cost per task.
- ChatGPT Work and Codex access covers Plus, Pro, Enterprise, and Edu users. Enterprise and Edu workspace admins must enable the models.
What it means Start with Sol for everyday coding before paying Astra’s higher rate. The small gap in OpenAI’s coding test doesn’t make Astra the obvious default.
Source: OpenAIDevs
Devin Bots keep coding projects moving
Cognition’s Devin Bots can manage a coding goal across multiple sessions, starting agents to work on individual tasks and retaining context between them.
- A bot can watch a schedule, Slack channel, or Linear board, then start separate Devin sessions and review their work.
- In one Cognition case study, an engineer gave a bot responsibility for keeping Android aligned with iOS. The bot found blockers, filed work, and started sessions for tickets.
- Bots are available to existing Devin customers. Cognition hasn’t published separate pricing beyond its existing plans.
What it means Give a bot a specific responsibility with a clear way to check its work. Cognition’s Android example shows why that can go further than repeatedly assigning individual tickets.
Source: automater.ai
Google AI Studio adds data sync to Android apps
Android apps built in Google AI Studio now support Firestore data storage and authentication. You can build an app that signs users in and syncs their data across devices.
- The Android emulator in AI Studio now supports HD resolution and audio.
- US users can publish apps to their personal Play library in a few clicks.
What it means For account-based Android apps, the new data support is a bigger reason to choose AI Studio than the emulator upgrade. The US Play option still stops at your personal library.
Source: GoogleAIStudio
Quick hits
- A rush to patch AI-found vulnerabilities caused outages, a journalist reports, citing sources who say Wall Street executives and the NSA told Congress.
- Microsoft released MAI Code 2 and Code 2 Flash in GitHub Copilot and Foundry, adding its own coding models to the Copilot model picker.
- Oregon sued OpenAI over alleged child-safety failures, seeking stronger safeguards and at least $2.035 billion in damages.
- OpenAI named TensorWave as a provider in its AMD compute deal, with 457 megawatts of capacity expected to begin coming online in the first half of 2027, Fortune reports.
- Perplexity rolled out Multiagent for Max subscribers on the web, letting a lead agent divide work among models from different providers.
- OpenRouter added management key controls on all plans, letting you make a key read-only or limit it to one workspace.
GitHub repo of the day
SkillEvaluator: check whether your agent can load a skill
- What it does: NVIDIA’s tool checks whether an agent skill is likely to load in different coding tools. That can help you find a misplaced skill folder or an incompatible description before relying on it.
- How to start: With Python 3.12 to 3.14, run pip install skillevaluator, then skillevaluator harness-check ./my-skill. This check needs no API key or agent run.
AI workflow of the day
How to find selling points you can support
Shared by God of Prompt (@godofprompt) · View original post ↗
If your product page relies on claims like “high quality,” this three-prompt workflow helps you find facts you can use in your messaging.
- Gather your product page, specifications, pricing, reviews, and any competitor information. Run the three prompts in order in one AI chat.
- Ask for a fact inventory:You are a product analyst. From the materials below, list every fact about the product: specs, features, prices with their conditions, guarantees, claims. Put each in a table with its source and an evidence status: verified (test, certification, independent data), brand claim (we say so), or missing evidence. Mark words like 'best' or 'leading' as brand claims. Never invent a fact, a number or a competitor comparison. End with a list of what's missing. Materials: [paste]
- Ask for selling points from the inventory:Using only the verified facts from the table, find up to 3 selling points. For each one give me: the fact, the job the customer is trying to get done, the result they get, what we're comparing against, and one line of messaging. Drop anything that rests on a brand claim. If you have no competitor data, don't use 'unique' or '#1'.
- Ask it to rank the options:Rank these selling points. Put first the one that removes the biggest reason people don't buy. Real customer data beats our opinion. If two are too close to call, keep both and tell me the one test that would decide it: a lost-sale call, a split test or a competitor check.