The latest AI model releases, and how to run a new one the day it ships

Snapshot of the most recent frontier drops

A photograph, not a clock. Treat the dates as the floor: by the time you read this there may be newer checkpoints on the trackers above.

The question the trackers skip

Reading that Gemini 3.5 Flash is 4x faster, or that GPT-5.5 Instant is the new default, is interesting for about thirty seconds. The thing that decides whether the release matters to you is mechanical: when a new model id exists, how many steps stand between you and running it inside the tool you already work in? For most coding agents the honest answer is "wait for the next app release," because the model names are baked into a dropdown.

I build fazm, a native macOS app that wraps Claude Code and Codex through the Agent Client Protocol, so I had to make a decision about this on the runtime side. The decision was to never hardcode the model list. Below is exactly how that plays out when a model lands.

Three levers that decide your day-one latency

The part you can verify in the source

This is the uncopyable bit, because it is a property of the code rather than a marketing line. fazm is open source, so you can open the files and check.

The model menu is assembled by ShortcutSettings.updateModels() in Desktop/Sources/FloatingControlBar/ShortcutSettings.swift. It reads the list the ACP SDK reports via onModelsAvailable after a new session is created and rebuilds the picker. The only hardcoded entries are three fallbacks used before the agent answers:

// ShortcutSettings.swift
static let defaultModels: [ModelOption] = [
  ModelOption(id: "haiku",           label: "Scary (Haiku, latest)"),
  ModelOption(id: "sonnet",          label: "Fast (Sonnet, latest)"),
  ModelOption(id: "claude-opus-4-8", label: "Smart (Opus, latest)"),
]

The endpoint override is a single setting. In SettingsPage.swift it is an @AppStorage("customApiEndpoint") string. When you set it, ACPBridge.swift validates the URL and exports it to the agent process as the standard ANTHROPIC_BASE_URL environment variable:

// ACPBridge.swift
if let raw = defaults.string(forKey: "customApiEndpoint")?
     .trimmingCharacters(in: .whitespacesAndNewlines), !raw.isEmpty,
   let endpoint = Self.validCustomAPIEndpoint(raw) {
  env["ANTHROPIC_BASE_URL"] = endpoint
}

And the three backends are pinned dependencies in acp-bridge/package.json: @agentclientprotocol/claude-agent-acp@0.29.2, @zed-industries/codex-acp@0.12.0, and @google/gemini-cli@^0.42.0. Bumping one of those is what surfaces a new model id from that lab, no full app re-release required.

Backends merged into one model picker

When the new model is not the upgrade you wanted

A fair counterpoint: most days the latest release will not move your work much. For everyday coding and desktop automation, the loop around the model tends to matter more than the checkpoint. Whether your session survives a Mac restart, whether the context gets silently compacted, and whether you can fork a thread cleanly all change your throughput more than swapping one frontier model for another. That is the actual reason to care about runtime mechanics: not so you can chase every release, but so that when a release genuinely helps, there is zero friction to adopting it, and when it does not, you have lost nothing.