In this Agnes AI review, here’s what we’re going to cover: what Singapore’s new model actually is, how Agnes 2.5 Flash performed when I ran it through real builds, which claims are self-reported, and who should bother. Short version: it’s genuinely free, it genuinely builds — and the benchmarks need a pinch of salt.
Short answer
- Agnes 2.5 Flash from Sapiens AI (Singapore) is fully free — the whole model, no usage cap.
- Agnes Code builds from plain English: my landing page, a 3D game and a calculator all worked from single prompts.
- Benchmarks look frontier-adjacent (~82.7 SWE-bench Verified for Pro) — but they’re internal, so hold them loosely.
- 500K+ devs used the last version — this is a real third player, not a toy.
What Agnes is
Agnes is an AI model family from Sapiens AI, Singapore’s own homegrown model company. For years the AI race has felt like a two-horse thing — the United States on one side, China on the other. Agnes is a third name in the room, and it isn’t a toy: Singapore has a national plan to be a builder of AI rather than just a user, and this is that plan made real.
The team built for text, images and video from the ground up — a genuinely multimodal family. What just shipped is Agnes 2.5 Flash, alongside a desktop app called Agnes Code, with a bigger Agnes 2.5 Pro coming for the heavy jobs.
The headline that earns the review: 2.5 Flash is free. Not stripped-down free — the full model, with no cap on usage. In one recent week their models handled trillions of words of work without anyone getting a bill.
Agnes Code: the part that makes it useful
Agnes Code is a desktop app for agentic coding. You type what you want in plain English and it works inside your real project: reads your files, writes across many files at once, runs commands and checks its own work.
The previous Flash model already had over 500,000 developers using it, so this isn’t an empty launch — there’s a real crowd building with it. And because the interface is plain English, you don’t need to be a hardcore coder: if you can describe what you want, you can get somewhere.
🔥 Want this set up without the guesswork? A free frontier-chasing model is only useful if you actually build with it instead of watching demos. Inside the AI Profit Boardroom you get the Agent OS as a ready-to-install file, a 30-day roadmap, daily tutorials the same day new tools ship, and four live coaching calls a week where you share your screen and get unstuck. → Get access here
The benchmarks — and why to hold them loosely
| Test | Score |
|---|---|
| SWE-bench Verified (2.5 Pro) | ~82.7 |
| Multilingual coding (Pro) | ~78.7 |
| SWE-Atlas | Flash’s biggest jump over its predecessor — ahead of GLM 5.2 and DeepSeek V4 Pro on parts of it, per their own chart |
The honest bit: these are Agnes’s own internal tests, run by Agnes. They are not independent outside checks yet. Treat them as a strong signal, not a final scoreboard — the real benchmark is always your own work.
Those numbers sit in the same neighbourhood as the top models people talk about daily, which for a free model from a small country would have sounded wild a year ago. Now it’s just another Tuesday — and that’s the actual story: the gap between the giant labs and everyone else keeps shrinking.
My tests for this Agnes AI review
- A landing page from one prompt. I described the AI Profit Boardroom page I wanted — the weekly coaching calls, daily tutorials, prompt libraries, member map — and it read the project, wrote the files and laid the whole page out, a section per part. That’s work I’d normally hand off and wait days for.
- A 3D snake game. A deliberate stress test: movement, three dimensions, many moving parts from a single prompt. If a free model holds something fiddly like that together, the everyday pages and tools are well within reach.
- A funnel calculator. A single clean page that takes a few inputs — people in, calls booked, follow-through — and does the maths. A real planning tool from one plain description, no spreadsheet wrestling.
Pattern across all three: one prompt in, a working first draft out that I could shape from there.
Who it’s for, and four tips
Builders first — developers and makers turning ideas into working pages and tools without a big setup. But the plain-English interface widens that door considerably.
- Start small. One clean page or one small tool before your dream project — learn how it thinks.
- Be specific in plain words. List the sections, name the parts, spell out the inputs.
- Always check the work. Fast and free doesn’t change who’s in charge. Read what it made before you use it.
- Don’t chase every model. Pick one, actually build, get good — that beats collecting fifty tools you never open.
FAQ
What is Agnes AI?
A multimodal model family from Sapiens AI in Singapore — text, images and video — whose new Agnes 2.5 Flash is fully free with no usage cap, alongside the Agnes Code desktop coding app.
Is Agnes 2.5 Flash really free?
Yes, and not trial-free: the full model with no cap, and Agnes says that isn’t ending on a hidden date. Their models handled trillions of words in a single recent week without bills.
How good are the benchmarks?
Strong — ~82.7 on SWE-bench Verified for the Pro model — but they’re internal tests run by Agnes themselves. Treat them as signal, not scoreboard, until independent checks land.
What is Agnes Code?
A desktop app where you describe what you want in plain English and it works in your real project: reading files, writing across many at once, running commands and checking its own output.
What did it actually build in testing?
A full landing page from one description, a 3D snake game as a stress test, and a working funnel calculator — each from a single plain-English prompt.
Who should use it?
Builders first, but anyone who can describe what they want clearly. Start with one small page, be specific, and always review the output.
The Agnes AI review verdict
Agnes AI review verdict: a genuinely free, no-cap model from Singapore that builds real things from single prompts, with benchmarks worth respecting but not yet worth quoting as fact. The bigger signal is the direction — frontier-level AI that doesn’t have to be expensive or imported — and the smart move is the boring one: test it on your own work.
About Julian Goldie
I run Goldie Agency, a 7-figure SEO agency, and teach this daily on a 400K+ subscriber YouTube channel. 240+ client projects on Upwork at a 100% job-success score, 10+ years through every major Google update. My systems are in the AI Profit Boardroom; my link building book is free here.
Related reading
Last updated August 2026. This is the living guide to Agnes AI review — it gets updated as the tools change.