changelog

Small. Weekly. Honest.

Every change we ship, in reverse-chronological order. If we're not shipping, this page will make that obvious.

15 releaseslatest 1.9.0

latest release

1.9.0 · Nov 24, 2026

Voice attempts across all subjects

Voice attempts now supported on every subject, not just math.
Whisper-based transcription with 0.4s median latency.
Recall queue now surfaces cards where you struggled with the verbal reasoning specifically.

2026 · 14 releases

2026

1.8.2

Nov 10, 2026

Study groups out of beta

Study group mode is generally available on Student and Parent plans.
Groups up to 6 (previously 4).
Session moderator can pin an Explain-back for group review.

1.8.1

Oct 27, 2026

Subject-level Brain score

Brain score now decomposes by subject.
New /app/score view with a per-subject trend line for the last 12 weeks.
Weekly cron rewritten to compute per-subject in one pass.

1.8.0

Oct 13, 2026

Native mobile web

Full mobile PWA support.
Camera capture works from Learn mode's attempt field.
Recall queue supports one-thumb review with keyboard shortcuts for anyone using a phone with a Bluetooth keyboard.

1.7.4

Sep 29, 2026

Streaming tutor replies

Tutor replies now stream token-by-token.
Output guard runs in parallel with streaming; if it detects a leak mid-stream, the visible reply is replaced with the refusal template with a subtle fade (about one in two hundred replies).

1.7.3

Sep 15, 2026

Recall pause

New 'pause recall' setting for exam weeks.
Pausing preserves your intervals so cards don't collapse.
Automatic resume after 14 days if you forget.

1.7.2

Sep 1, 2026

Better Explain-back grading

Rebuilt the Explain-back grader prompt after 12 weeks of user feedback.
Grader now returns targeted feedback that points to the specific weak link in the explanation.
Average grader satisfaction rating up from 3.9/5 to 4.5/5.

1.7.1

Aug 18, 2026

Classifier tuning

Retuned the Efficiency-mode classifier with 3,000 real user messages as training signal.
False-positive rate down from 8% to 3%.
False-negative rate flat at 4% (well below the fallback tutor safeguard).

1.7.0

Aug 4, 2026

Image attempts

You can now submit a photo of handwritten work as your attempt.
Sonnet vision transcribes and reasons over the image alongside any text you include.
Math and chemistry beta users have been asking for this since day one; here it is.

1.6.4

Jul 21, 2026

Cheaper hints

Hint cost reduced from 15 to 10 Brain points.
Daily regen up from 40 to 50, cap at 200 (previously 150).
Economy tuning driven by data: too many users were rationing hints instead of using them.

1.6.3

Jul 7, 2026

Parent dashboard refresh

Parent dashboard is now week-based instead of daily.
Sub-account switcher (for the Parent plan).
Effort trend line to see whether the kid's practice is compounding.

1.6.2

Jun 23, 2026

Confidence auto-nudge

If your calibration score is systematically over- or under-confident, the confidence picker now shows a subtle hint before you commit.
Not shown to users whose calibration is above 80.

1.6.1

Jun 9, 2026

Faster recall UI

Recall queue interactions now feel instant.
Optimistic UI everywhere; server error rolls back with a soft toast.
Median review time down from 22s to 9s.

1.6.0

May 26, 2026

Subject-scoped Learn sessions

New Learn sessions default to your subject with fewest recall reviews this week.
You can override, but the nudge helps most users.

1.5.4

May 12, 2026

Prompt caching everywhere

All system prompts now use Anthropic's prompt caching.
Per-problem context also cached, so multiple turns on the same problem share the cache.
Total inference cost down ~45%; savings passed through the free tier limit going from 10 to 15 messages a day.

if a week is missing, we shipped nothing worth writing down

start now