Updates

Changelog

Version history of the pro-1 model and platform changes. The service is in early access — improvements ship frequently.

pro-1 model versions

pro-1-2026-07-06 currentImmutable snapshot of the production PRO-1 behavior. Use pro-1 for the continuously improving release.

11 July 2026 · Pricing update

PRO-1 text pricing changed to $12 / 1M processed input tokens and $48 / 1M processed output tokens.
Pricing version pro-1-usd-2026-07-11-r2 identifies the new tariff independently from model behavior versions.
The public pricing endpoint, landing, documentation, account workspace, estimates, model cards and billing metadata were updated from one runtime plan.

11 July 2026 · Pricing (superseded)

An earlier 11 July decision set text rates at $15 / $30 (version pro-1-usd-2026-07-11); it was superseded the same day by the $12 / $48 tariff above (pro-1-usd-2026-07-11-r2), which is the production rate.
Added GET /v1/pricing as the machine-readable tariff contract.
Landing, documentation, estimates, model cards and response billing metadata render from one runtime pricing plan.
Cost examples use explicit processed-token counts instead of an understated typical-request promise.

10 July 2026 · Platform

One public model namespace: use pro-1 for the continuously improving model or pin an immutable pro-1-YYYY-MM-DD snapshot.
Pricing is prepaid pay-as-you-go only, with no subscriptions or promotional credits.
The benchmark section now centers the current Hard-24 no-regret result; the pre-no-regret Production-60 slice was retired from the current public comparison.
Documentation, account onboarding, realtime tools and model discovery now use the same single-model contract.

29 June 2026 · Platform

Voice mode (realtime) and an image & video generation studio in Chat.
Usage tiers 1–5 with request limits for load control; your tier and upgrade terms are in your account.
Full data isolation: chats, sessions and files are strictly private to your account.
Account: 30-day spending history, password change, API token masking, low-balance warning.
Chat was reworked with saved conversations on the left.
Reading your own history and session status is available even at zero balance.

28 June 2026 · Platform

Multimodal input: images (analysis), audio (speech recognition), files and PDF (text extraction).
Open machine-readable benchmark proofs: /benchmarks and /benchmarks.
Streaming responses (SSE) with a neutral progress block.
Billing by processed tokens — standard $/1M format (input $15 / output $30), with a pre-run cost range via /v1/estimate.