Last week, Moonshot AI released Kimi K3. 2.8 trillion parameters. The largest open-weight model ever announced.
It scored above Claude Opus 4.8 and GPT-5.5.
When the weights drop, anyone with a GPU rig runs a near-frontier model for almost nothing. The market responded with $3.3 trillion in chip-stock losses.
Investors asked the obvious question: if a model this good runs almost anywhere, what is all that infrastructure spending actually buying?
But there's a quieter question underneath. And it's the one that actually matters to you.
Why Model Loyalty Is Losing
Not long ago, picking a model was straightforward. You chose the one everyone trusted, integrated it, and moved on. That worked when one model stayed on top for years.
That era is over.
GPT was the default until it wasn't. Claude took the crown for reasoning until Kimi K3 arrived and rewrote the benchmark. Whatever you standardise on today gets undercut, out-benchmarked, or open-sourced by the next release.
The model you bet on isn't getting worse. Something better, cheaper, or more open always ships next.
Betting your entire workflow on one vendor's roadmap means betting you keep losing.
The Model-Agnostic Difference
What does model-agnostic actually mean? It means you stop choosing a model and accepting every consequence that follows.
Instead, you describe the task. Euryka evaluates which model does that job best, at the price you're willing to pay. Kimi K3. Fable 5. GPT 5.6. Grok 4.5. Whatever ships next month.
Your team doesn't track benchmark leaderboards or chip-stock headlines to know they're using the right tool. That's infrastructure's job. Not yours.
Five New Models Arrive
This release adds five models to your available lineup, all live now in the model selector.
GPT 5.6 comes in three variants — Sol, Terra, and Luna — each tuned for different reasoning styles and conversation depths. Grok 4.5 arrives for reasoning-heavy work. Muse Spark 1.1 rounds out the general chat options.
Alongside these additions, Kimi K3 and Fable 5 are now integrated and ready to use. Pick them directly, or let Euryka route based on task fit and cost.
Smart Cost Warnings Before You Commit
Premium models cost real money. Fable 5 is exceptional. It's also overkill for most jobs.
When you switch to a high-cost model, Euryka warns you first. It shows the price difference and a lower-cost alternative so you can decide whether the premium capability is worth it before it applies.
No surprise credit burn. No accidental overspend. You choose the expensive option when the task actually needs it, not by default.
One Media Browser Across Every Tool
Wherever you need to add, remix, or reference media, the same consistent browser now appears — tuned to whatever tool you're in.
Pull an existing asset. Reuse a past generation. Bring in a reference. All without leaving your workflow. One unified way to find and reuse images, audio, and video across the entire platform.
Domain Health Tracking for Prospects
Teams running outreach at scale now have visibility into domain reputation before it becomes a problem.
A new DomainMetrics panel on your Prospects domain pages shows a rolling 30-day view of send attempts, deliverability, and bounce rates — split into transient, permanent, and failed. Catch domain issues before they tank your sender reputation.
Luma Ray-2 Now Handles Longer Videos
Luma Ray-2 reframe used to reject anything over 10 seconds. Now it chunks longer sources automatically, reframes each piece, and stitches them back together with the original audio intact.
You can now reframe sources up to 60 seconds. Full-length video in, reframed video out, no manual stitching required.
Two Fixes Worth Noting
On-demand spend limits now reset correctly each billing period, instead of staying blocked after a single overage. And the app installs properly as a standalone PWA — no more login redirects bouncing you around.
Why This Release Matters
Models will keep leapfrogging each other. That won't stop.
Your job isn't to pick the winner. It's to build on infrastructure that lets you switch the moment the winner changes.
That's what Release #29 is about. Not a single new model that becomes your standard. Five new options, smart routing, and a system designed to stay current as the landscape shifts.
Stop backing a model. Back a system instead.
How We Got Here
Each release built toward this moment. Multi-Model Fan-Out let you generate with several models at once and compare results side by side. Before that, we made your daily conversations sharper with inline media and @ mentions. And we gave you a visual canvas for designing entire campaigns once and running them forever.
