Gate 15 offers four frontier model families — Anthropic Claude, Z.ai GLM, OpenAI GPT, and Google Gemini. Auto routes across Claude and Gemini; GPT and GLM remain available as manual choices for any turn.
Don’t want to think about which model to use? Auto is the default. Gate 15 reads what you’re asking for and routes it to the model best suited to the moment — a careful planner for mapping out a change, a strong coder for building it. You can always override per turn or set your own default model in Settings, and switch mid-project whenever the work changes shape.
The default for every new project — zero setup
Routes planning and building to the best-fit model
Override any turn, or set an account-wide default
Four providers, one workspace
Four strong models, each great at something.
They each have a personality. Knowing the differences helps you reach for the right one; Auto currently chooses between Claude and Gemini.
Anthropic
Claude (Opus / Sonnet)
The careful builder. Reads the whole picture before it moves, lays out a plan, and makes precise, well-reasoned changes across a project.
Deep planning and careful, surgical edits
Long, focused building sessions that stay on track
Clear explanations of what it changed and why
Z.ai — GLM
GLM-5.2
The value pick. Near-Opus coding quality at a fraction of the cost, with a 1M-token context that swallows whole codebases in one go.
Near-Opus coding quality for a fraction of the price
1M-token context for large, sprawling projects
Strong, cost-effective everyday building
OpenAI
GPT-5.5 / GPT-5.3 Codex
The problem-solver. Strong at writing code, untangling tricky bugs, and pushing through multi-step tasks without losing the thread.
Complex problem-solving and debugging
Fluent, idiomatic code across languages
Sticks with long, multi-step tasks
Google
Gemini
The fast synthesizer. Built for working through a lot of information at once — great for research-heavy work and quick turnarounds.
Working through large amounts of context
Research-heavy tasks and broad synthesis
Fast turnarounds when you want momentum
Dial in how hard it thinks
Thinking effort is a per-turn control over how much the model reasons before it answers — a full slider from low up through max, scoped to whatever rungs the current model actually supports. Set it right in the composer for a single turn, or pick an account-wide default in Settings. New projects start on medium, which fits most building work. Reach for high (or beyond, up to max on models that go that far) when a task is genuinely hard, or low when you just want a quick pass.
Per-turn control, set in the composer’s model picker
Account-wide default in Settings — defaults to medium
Works across all four providers, each scoped to its own range
LowQuick passesSnappy answers for small edits, copy tweaks, and questions where you already know roughly what you want.
MediumThe defaultA balanced amount of reasoning that fits most building work — careful enough to get it right, fast enough to keep moving.
HighHard problemsMore room to think through tricky rebuilds, gnarly bugs, and decisions where one wrong turn costs you an afternoon.
X-High / MaxGo all inThe deepest rungs, for the models that support them — for when you want the model to spare no reasoning before it acts.
Native web search across providers
When the answer depends on something current — a library’s latest API, a framework that shipped last week, today’s docs — the model can search the web on its own. Capped Claude search works with Gate 15 credit. GLM, GPT, and Gemini 3.x use their native search when you provide that provider’s key, so an unbounded provider tool can never silently spend the shared wallet. Every search shows up in project activity, so you can see exactly what it looked up.
Capped Claude search works with shared or personal credentials
GLM, GPT, and Gemini search is available with your provider key
Used only when the answer needs fresh information
latest framework release notes
Anthropic Claudeweb search
Z.ai GLMweb search
OpenAI GPTweb search
Google Geminiweb search
Side by side
How the providers compare.
Every model in the picker is a curated, full-strength choice. We leave out the lighter tiers — Haiku, Flash-Lite, mini and nano — so you never have to wonder whether you picked one that’s too small for the job.
Capability
Claude
GLM
GPT
Gemini
Built-in web search
✓
with your key
with your key
3.x + your key
Thinking effort (low → max)
✓
✓
✓
✓
Best for
Deep planning & careful changes
Great coding value & huge context
Complex problems & multi-step code
Research & fast turnarounds
Works on Auto
✓
—
—
✓
All four are always available to pick yourself, per turn or as your account default — Auto currently chooses between Claude and Gemini automatically, with GLM and GPT support on the way.
A curated list, not a menu of trade-offs
New models land constantly, and most of them aren’t worth your time. Gate 15 keeps a hand-picked, single source of truth for the models you can select, shared by the workspace and the router so the choice you see is always the choice that runs.
We deliberately exclude the low tiers — Haiku, Flash-Lite, and the mini and nano variants. They’re cheaper, but they cut corners on exactly the kind of careful, multi-step building this product is for. The models you can pick are the ones we’d pick. If you’d rather not think about it at all, leave it on Auto and let Gate 15 choose for you.
AI app builder
Ready to build?
Start with $3 in trial usage. BYOK from $8/month. Included model credits from $20/month.