Foxl v0.4.22: Claude Opus 5 Is the New Default

Claude Opus 5 is Anthropic's most intelligent model, and as of today it is the default in Foxl Agent, Foxl Code, and Foxl Notes. Not an option buried in a dropdown - the default, on day one, at the same price we were already paying for Opus 4.8.

Why it becomes the default and not just another choice

A new flagship usually arrives with a tradeoff: better results, higher bill, so you reach for it deliberately and use something cheaper the rest of the time. Opus 5 does not have that shape. It matches or beats Fable 5 on agentic coding and professional knowledge work at half Fable's list price, and it lands on exactly the same $5 in / $25 out per million tokens as Opus 4.8, which it supersedes. There is no version of "stay on 4.8 to save money" that makes sense, so we did not make you choose.

It carries a 1M-token context window and up to 128K tokens of output. Reasoning is on by default: rather than you picking a thinking budget, Opus 5 decides for itself how long to think on each turn, and the effort control (low through max) sets the depth rather than switching it on. If you had reasoning configured before, your setting still applies.

One consequence worth stating plainly, because it broke things internally before we caught it: on Opus 5 a request thinks even when you did not ask it to, and the reasoning block can arrive before the answer text. Any code that grabbed the first block of a response and assumed it was the reply now gets a thought instead. We found and fixed two places in Foxl Notes where that would have produced silently empty summaries.

Your model list stops being an archive

Every model we have ever shipped was still in every picker. Opus 4.7, Opus 4.6, Sonnet 4.6, GPT-5.4 and older, Gemini 2.0 - all sitting in the same list as the models you should actually pick, in a fixed-height scroller. The list had become a changelog.

Superseded models are now hidden by default across all four pickers: chat, Settings, Foxl Notes, and Foxl Code. Nothing was removed. A single switch - "Show older model versions" in Settings, or the one-tap row at the bottom of any picker - brings them all back, and it is one preference shared by all three products rather than four that drift apart.

The rules for what stays visible matter more than the hiding itself. A model you have selected is never hidden, because a picker that hides the model you are currently running just shows you an empty box. A model you have favorited is never hidden, because pinning something is a stronger signal than our tidiness default. And search always finds everything: type "4.6" with the setting off and Sonnet 4.6 is right there. Hiding is a default, not a restriction.

Haiku 4.5 stays visible, for the record. It is the current fast tier, not an old version.

Things that were quietly broken

Auditing every model engine for stale configuration turned up a set of bugs that had nothing to do with old versions and everything to do with nobody checking.

Gemini CLI returned an empty reply to every single request.Foxl read the CLI's output stream in a format that CLI has never emitted, so every line of every answer was discarded and you got a blank bubble. Errors were worse than silent: a quota or sign-in failure was indistinguishable from success, because the field we checked for "did this fail" is one the CLI never sets. We installed the shipping CLI and read its own event definitions out of the bundle to get this right rather than trusting documentation. On the way we also found that signing in with gemini was never detected at all - Foxl only looked for an API key, so a signed-in CLI still reported itself as not configured.

Some models were not the model you selected. Choosing the short "Claude Code Sonnet" option ran Sonnet 4.6 long after Sonnet 5 shipped, because its sibling Opus alias had been updated and this one had not. Six other model identifiers resolved to nothing internally, which made Foxl treat a 1M-context model as though it held 256K - compacting your conversation about five times earlier than necessary - and record its cost as zero. GPT models running on a ChatGPT subscription were assumed to hold 1M tokens when the service actually serves 272K, so long sessions failed upstream instead of compacting in time. A GPT-5.3 option that OpenAI does not serve at all has been removed.

Foxl Code was erasing your starred models. Foxl Code and Foxl Agent share one favorites list but stored it in two different formats, so opening Agent chat once silently unstarred everything you had favorited in Code.

Our own pricing documentation was wrong by 9x. The per-token credit table in the docs assumed one credit buys four cents of model usage. It buys $0.036 - there is a margin, and the table never accounted for it, so every row understated the real cost by a factor of 9.26. Both tables are recomputed from the actual rate, and the rate is now stated on the page so the arithmetic can be checked. We would rather publish a correction than leave a number that looks reassuring and is not.

Also in this release

Settings gained a Refresh button next to Save Provider Settings: after running claude /login or codex login outside the app, Foxl used to keep showing the old "not configured" state until you restarted it. Foxl Code now bills AgentCore compute properly - a task costs both the model tokens and the machine it runs on, and the machine half was never charged on relay-routed tasks. The Budget section lost two knobs that did nothing at all, and Runtimes now shows real live spend instead of $0.000 for the entire duration of a task.

On mobile, the starter prompts under the greeting can finally show something (they never could), your last greeting is remembered so a cold start paints instantly, and tapping a product in the sidebar closes the drawer like every other destination in it. The chat send button is round now, matching Foxl Code.