OpenAI says its models account for most of the text processed by Amp. Amp’s Quinn Slack says model improvements now let its AI assistant help developers decide what to build, write code and design pages. Some Amp users process 10 billion tokens—a measure of text used by AI models—in a week, Slack says. That makes access to model capacity a practical software-delivery question, not just a technical one.
Amp has launched Sign in with ChatGPT. Slack says the integration gives Amp users access to more tokens and OpenAI models, including GPT‑6 Astra. The important connection is between a wider job description for the assistant and the capacity to do that work. For a business commissioning software, the next question is whether a vendor’s AI assistant can carry a project from planning through a usable page, rather than merely produce code.
Why It Matters for Business Leaders
Why it matters: When buying software work, ask to see the finished page and the decisions behind it—not just a demonstration of how quickly an assistant writes code.
Ai2 says HiRO-ACEv2 can generate global climate data at three-kilometer detail for about one-300th the cost of X-SHiELD. Ai2 also says HiROv2 reproduces extremely rare rain, wind and pressure events with high accuracy.
In Plain English: Producing a detailed weather-risk map could become less like commissioning an expensive one-off study and more like updating a working map as questions change.
Why it matters: Businesses exposed to floods or severe storms can ask insurers and risk advisers whether cheaper simulations would improve location-specific planning.
■ The Street Says
The mood is split between excitement about what models can create and impatience with the practical limits of access, cost and research computing.
↳ Re: @OfficialLoganK’s conversation with Sergey Brin and colleagues about Gemini 4 Argon.
“the creativity in the prompting and the model is awesome to see”. For business owners, judge creative AI by the work it helps people produce, not a model name alone.
↳ Re: Embedding Gemma 2 by Google appearing on Hugging Face’s trending list.
“The media and silicon valley bubble only talks about reasoning LLMs”. Look beyond headline-grabbing assistants when choosing AI; tools for finding information or handling audio may fit the job better.
↳ Re: a researcher’s wait for university computing capacity.
“it takes 3–4 business days just to START running”. If your team depends on scarce computing capacity, count queue time when setting a research deadline.
↳ Re: Fish's Drama 3 35B3A running on a DGX Spark.
“The 4bit quants get very close to the original”. For a voice project, compare samples from a smaller version before assuming the largest version is worth its cost.
Research teams can give an AI assistant a more structured starting point for scientific work.
What it does: This library offers 177 ready-to-use, validated skills and access to more than 100 scientific databases.
Problem it solves: It reduces the setup work of finding reusable scientific tasks and relevant data sources separately.
Who should use it: Research and development teams building AI-assisted scientific workflows.
Expected outcomes: A shared collection of scientific tasks and database connections that a team can use as the starting point for its own work.
■ Two Steps Ahead
Over the next 30–90 days, Amp’s Sign in with ChatGPT could make model-access allowances as important as the coding assistant’s features. When comparing software-development proposals, separate the cost of the assistant from the capacity available to its users.
HiRO-ACEv2 gives weather-exposed businesses a specific question for their next insurance or site-planning discussion: could more detailed simulations change which locations deserve attention? For research teams, scientific-agent-skills could shorten the work of assembling scientific tasks and data sources, without turning a community library into an enterprise purchase decision.
Tools & Platforms to Watch
Amp: Help a software team plan, write and design parts of a project.
scientific-agent-skills: Give a research team reusable scientific tasks and database resources.
