Free Newsletter
Get the hottest Fintech Singapore News once a month in your Inbox
Agnes AI has launched Agnes-2.5-Flash, the newest generation of its text model, purpose-built for coding and agentic tasks.
The company also announced Agnes-2.5-Pro, a larger flagship model aimed at more complex, long-running agent work, which follows later this month.
Alongside the model release, Agnes AI introduced Agnes Code, a desktop workspace where developers can put the new model to work directly inside real projects.
Bruce Yang
“The constraint on agentic coding isn’t capability. It’s cost. Agnes-2.5-Flash removes that constraint entirely.
Agnes-2.5-Pro pushes further for teams that need more depth.”
said Bruce Yang, Founder & CEO of Agnes AI.
Coding agents got more capable. They also got more expensive.
The shift now happening across the industry is toward agents that run for extended stretches on their own, planning tasks, editing across multiple files, executing commands and checking their own results, rather than answering one prompt at a time.
For developers who have adopted this workflow, it represents a fundamental change in how software gets built.
That shift has a well-documented downside: as usage deepens, so does the bill.
Developers running agents daily are hitting usage limits and ballooning costs precisely when the workflow becomes indispensable, a tax on exactly the behavior the industry is trying to encourage.
Agnes-2.5-Flash is built to remove that tax, not manage it, so developers can run agents as heavily as their work demands without watching a meter.
Agnes-2.5-Flash: built for how developers actually use agents
Agnes-2.5-Flash is a direct successor to Agnes-2.0-Flash. The improvements show up most in the tasks agentic workflows depend on:
- Generating code across multiple files
- Holding project context over a long task
- Calling tools correctly
- Recovering when a step in a multi-turn task doesn’t go as planned
The underlying agent orchestration layer, the Agnes Harness, has also been upgraded, improving how reliably the model carries out multi-step tasks without losing track of state.
Agnes-2.5-Flash is available now through the Agnes AI developer platform and API.
What developers were already saying
Developers using Agnes-2.0-Flash, already live via the API with more than 585,000 users, had flagged it as strong for real coding work, feedback the team built on for this release:
“I think Agnes Flash is suited perfectly for code completions. I might use it in VS Code for that. It’s strong for CSS already and solid across general coding,”
said one developer in the community.
“Pretty good TTFT. It’s pretty optimized,”
said another.
Agnes-2.5-Flash builds directly on the model these developers are describing.
Agnes-2.5-Pro: for the hardest tasks, later this month
Where Flash focuses on speed and everyday coding, Agnes-2.5-Pro is designed for more demanding work.
This includes production-grade systems, complex agent pipelines and tasks that require deeper reasoning over longer periods.
The model is aimed at professional development teams as well as solo builders, researchers and power users who need more than a fast model can provide.
Agnes-2.5-Pro follows later this month. Pricing and access details will follow closer to release.
Agnes Code: a workspace, not a chat log
For developers who want a project-based workspace rather than raw API access, Agnes-2.5-Flash is available today through Agnes Code, a new desktop app launching today.
Agnes-2.5-Pro will join the workspace when it becomes available.
Agnes Code lets developers describe a task in plain language and have it carried out directly inside a real project, reading file structure, editing across files and running commands, with results saved to the project rather than scattered across a chat log.
It runs on the same account, subscription and credits as Agnes AI’s web and API products, so a developer pays once and uses their balance everywhere, across chat, API and Agnes Code, instead of juggling separate plans.
Free access at production scale
Agnes AI’s free-access model isn’t new. What’s new is the scale it now operates at under the Agnes 2.0 model series.
Over the past seven days, Agnes AI’s text, image and video models processed 5.41 trillion tokens combined, including 3.25 trillion text tokens and 2.16 trillion multimodal tokens, under the existing free-access programme, with no usage cap and no announced end date.
For scale, 3.25 trillion text tokens in a week is on the order of the text in millions of full-length novels, sustained every seven days at no cost to the developers generating it.
That volume is the clearest evidence that removing cost as a barrier changes how people build. This isn’t casual experimentation.
It’s sustained production usage at a scale the infrastructure is already handling today. Agnes-2.5-Flash launches into that same free-access model, now extended to Agnes AI’s most capable coding release yet.
A preview, shaped by real usage
None of today’s releases are being treated as finished. Agnes AI is putting Agnes-2.5-Flash and Agnes Code in front of real developers early and shaping what comes next based on how they are actually used.
Agnes-2.5-Pro follows later this month, alongside updates to Agnes AI’s image and video models as part of the broader Agnes Full-Modality Model Series.
Get started
Featured image: Edited by Fintech News Singapore, based on image by Agnes AI
