Drevon v0.4.18 - Latest AI models, custom IDs, and safer fallbacks
0.4.18 is a model support release for the pace of AI tooling right now: fast, uneven, and a little chaotic around rollout windows.
Drevon now knows about the latest model options across GitHub Copilot CLI, Claude Code, and OpenAI Codex, including GPT-5.5 and Claude Opus 4.7. Just as importantly, it handles the messy parts better: plan-gated models, account-specific availability, custom model IDs, and fallback paths when a CLI says a model is not available yet.
What's new
Latest model support across all three CLIs
Drevon's model picker has been updated for the current generation of coding models.
For GitHub Copilot CLI, you can now select newer OpenAI and Anthropic options such as:
gpt-5.5gpt-5.4gpt-5.4-minigpt-5.3-codexclaude-opus-4.7claude-sonnet-4.6
For Claude Code, Drevon now supports both the practical aliases and the pinned model IDs:
sonnetfor the latest Sonnet lineopusfor the latest Opus lineclaude-opus-4-7claude-sonnet-4-6claude-haiku-4-5
For OpenAI Codex, the picker now includes:
autogpt-5.5gpt-5.4gpt-5.4-minigpt-5.3-codexgpt-5.3-codex-sparkgpt-5.2
The goal is simple: if your CLI account can use the model, Drevon should not be the thing holding you back.
Better defaults for new agents
New agents now use defaults that are more resilient to rollout differences.
Copilot and Codex agents default to auto, which lets the CLI choose the best available model for your account and plan. Claude Code agents default to sonnet, which follows Claude Code's latest Sonnet line while still allowing you to pin exact model IDs when reproducibility matters.
Existing agents are not silently migrated. If you already chose a specific model, Drevon keeps that setting. That means old agents stay predictable, while new agents get better defaults.
Custom model IDs
Every CLI moves at its own speed. Sometimes a model is available in your account before Drevon has shipped a new desktop build with that model in the dropdown.
0.4.18 adds a custom model ID escape hatch for exactly that case.
From the model menu, you can now enter a CLI-supported model ID directly. Drevon will preserve it, display it as a custom model, and pass it through to the underlying CLI. This works for Copilot, Claude Code, and Codex.
This should make future model launches feel much less blocked by app release timing.
Clearer fallback handling
Model availability is not just about the CLI version. It can vary by account type, plan, organization policy, geography, and staged rollout.
Before this release, if you picked a model that your CLI could not actually use, the failure could look like a generic agent startup problem. That was noisy and not especially helpful.
Drevon now watches for unavailable-model errors from the runtime. When it detects one, it surfaces a clearer message and offers a fallback model from the catalog. For example, a GPT-5.5 rollout issue can point you toward GPT-5.4, and a Claude Opus 4.7 availability issue can point you toward the previous Opus option.
This does not hide the underlying limitation, but it gives you a faster path back to work.
A cleaner model system under the hood
The visible part of this release is the model picker. The more important part is the foundation behind it.
Drevon now has a shared typed model catalog with lifecycle metadata: auto, recommended, stable, preview, fallback, legacy, and custom. The app uses that catalog for labels, defaults, fallback choices, and dropdown ordering.
That means future model updates should be smaller and safer. Instead of updating several separate hardcoded lists, Drevon now has one source of truth that both the UI and runtime can understand.
What this means in practice
If you use Drevon with Copilot, Claude Code, or Codex, 0.4.18 gives you more control over which model powers each agent.
Use auto when you want the CLI to make the best available choice. Use a recommended model when you want to try the latest capability. Use a pinned model when repeatability matters. Use a custom ID when your account gets access before the desktop app has a named option for it.
And if a model is not available yet, Drevon should now explain that more clearly instead of leaving you to guess whether the agent, auth, or CLI failed.
0.4.18 is a small release on the surface, but it makes Drevon's agent layer much more future-proof. Models will keep moving quickly. Drevon should make that pace easier to work with, not harder.