Models

Skydive separates the agent from the model that runs it. You can change the model at any time (choose from 300+ frontier and open source models). You pay 0% markup on provider API rates. For no-training and zero data retention settings, see Model Settings.

​

Choosing a model

​

Change the model on an agent

The model is set per agent. Switching updates that agent only. The new model (and thinking level, if you set one) applies on the agent’s next run.

1. Open the agent (from chat or the agent profile).
2. Open the model picker (the model name by the composer or agent controls).
3. Choose from the six featured models, or search to reach any model in the catalog.
4. Hover a model to see its intelligence, speed, and cost ratings and pick a thinking level.
5. Confirm the switch.

When you switch models inside a live conversation thread, the change pins to that thread alone. Your other open conversations and the agent’s default are untouched. If you open the picker on the new-chat screen before a conversation starts, the selection becomes the agent’s new default instead.

​

How the picker works

The picker opens with six curated models: Glide, Claude Opus 5.5, GPT-6 Sol, Grok 4.7, Gemini 3.8 Flash, and GLM 5.3 Flash. If your agents are running a model outside that set, it shows as an extra row until you switch. Rest on any model to see its intelligence, speed, and cost ratings (sourced from Artificial Analysis) and click a thinking level directly from the card. Search to reach any model in the full catalog. Models you pick from search save to your set and show in future sessions.

​

When your workspace is out of credits

If your credit balance is empty and Glide is serving as the free fallback, the picker adjusts:

• Glide shows a “free” tag so you can see exactly which model your agents are running at no charge.
• Non-Glide models are locked. The picker dims them and won’t let you select one until you add credits. In the TUI, the status bar shows “needs credits” while a locked model is highlighted.
• Search is replaced by a plan notice. Instead of a search box, the picker shows a short message explaining that some models are unavailable. Glide is still selectable normally.

Once you add credits, the picker returns to its normal state.

​

Match the model to the job

Reasoning-capable models expose a thinking level. Turn it up for harder problems; turn it down for speed and lower cost.

​

Cost-effective picks

Model usage is at cost. Provider rates pass through to your credit balance with no markup on model spend. Costs below are USD per 1M tokens (input / output), from the live registry.

​

Strong defaults

​

When you need the flagship

​

Thrifty and high volume

​

Skydive Glide

Glide is Skydive’s own model option. It delivers strong everyday performance across agent work (drafting, research, tool use, multi-step tasks) and it is the default for new agents when your workspace has not set a different default. A few things that make it different from picking a provider model directly:

• Skydive manages the target. Glide is an alias that routes to the best available model for the job at any given time. Skydive updates the target when a better option emerges, so your agents stay current without you changing anything.
• Quality and cost in balance. Glide sits at a mid-tier price point while targeting performance you would expect from a capable everyday model. Rates appear in the model catalog.
• No upcharge on any model. Whether you run Glide or pick a model yourself, Skydive passes provider rates through at cost. You pay the same rate regardless of which model is underneath.
• Fallback is built in. If the underlying model is unavailable, Glide routes to a backup at the same or lower price, never up to a premium model.

Glide is a solid default for most agents. Switch to a flagship when the work is genuinely hard: long reasoning chains, careful judgment, ambiguous briefs. Switch to a flash model when you are running high volume and the task is light.

​

Use your own provider or gateway key

Workspace owners and admins can connect their own API keys for billing, discounts, compliance, or provider account controls.

1. Open your workspace settings.
2. In AI providers, select Add provider.
3. Choose a provider or gateway.
4. Enter its API key, then select Connect.

Skydive verifies the key before saving it. The connection applies to the entire workspace. You can connect a gateway key for OpenRouter or Vercel AI Gateway. Skydive routes compatible models through your gateway account. Your gateway controls billing and any provider routing you configured there. You can also connect a direct provider key for Anthropic, OpenAI, Google AI Studio, xAI, or Z.ai. A direct provider key takes priority over a gateway key for that provider. Other compatible models continue through the gateway. To stop using a key, open AI providers in your workspace settings and select Disconnect.

​

Spend tips

• Default cheap, escalate one agent. Keep most agents on a mid or flash model. Only the agent doing the hard job needs the flagship.
• Turn thinking down when you can. Higher thinking levels burn more of your credit balance on reasoning-capable models.
• Set a workspace default. In Settings, Models, pick the default model so new agents start on the right tier without each one needing manual setup.
• Watch the Usage dashboard. See which models draw credit under Usage, and swap offenders to a thriftier pick.
• Prefer at-cost clarity over guessing. Rates in the catalog are the same pass-through rates billed to your credit balance.

​

How cost ties in

The model you choose affects both quality and cost. More capable models and higher thinking levels draw more from your credit balance per task. Model usage is at cost. Provider rates pass through with no markup. You pick the model for the job, not to dodge a hidden platform fee on model spend.