Changelog

v0.1.18 (2026-10-11): Some models may now have a limited-time price. The Models pages, the home page and Pricing show the regular price struck through, the limited-time price, how much is off and when it ends (Toronto time, ET), and the Playground shows it next to the model's rates (What is a limited-time price?). Times on your account pages now show in Toronto time unless you chose a time zone in Settings → Timezone.
v0.1.17 (2026-10-10): After signing in or creating an account you now return to the page you were opening — for example a Playground link with a model chosen — instead of always landing on your account overview. This works for email sign-in, Google and GitHub sign-in, and the link in the verification email. Model pages now describe their performance figures as the median of the last 10 test requests.
v0.1.16 (2026-10-09): The uptime on the Models pages and the home page now updates by itself: Uptime with a percentage and the period it covers — usually the last 3 days, with one block per day — measured from our own automatic test requests to the suppliers your requests use by default; some models may show a shorter period (the last 24, 12 or 6 hours) or the last 7 days, always named on the card. It is no longer a figure we publish by hand (How we measure uptime).
v0.1.15 (2026-10-09): Sometimes a model is temporarily available only from suppliers without a no-storage commitment, which the default Data handling setting does not allow. An API request for that model that fails now answers 503 model_limited_by_data_handling with a message pointing to the setting, and it is not billed (API reference). In the Playground the model shows "(limited suppliers)", and selecting it asks whether to use those suppliers for this session, billed at the list price (Playground guide).
v0.1.14 (2026-10-08): When our automatic checks cannot reach a model, the Models pages and the home page now show it greyed and marked Temporarily unavailable, its own page explains why, and the Playground does not offer it until it is back (Playground guide). The mark goes away by itself as soon as a check succeeds. The API still accepts requests for the model, and a request that fails is not billed (API reference).
v0.1.13 (2026-10-07): You can now limit how much of your balance is used beyond your Agent plan: on the Plan page, the card Balance use beyond your plan has a Limit per billing period (US$) next to "At a limit, continue on my balance", and shows what was used this period. When the limit is reached, requests beyond the plan get a 402 overflow_cap_reached with a Retry-After header until the plan pays again (how it works). On the Usage page, those requests carry a Beyond plan tag. The Plan page and the account sidebar warn you from 80% of the limit. The Agent plan's no-refund rule is now stated on the Terms, FAQ, Pricing and Plan pages.
v0.1.12 (2026-10-07): The OpenCode and WorkBuddy samples now show the live model's name and limits.
v0.1.11 (2026-10-07): Transactions now has two views: Payments lists what you paid — top-ups and every Agent plan charge (new subscription, monthly renewal, prorated plan change) with its Stripe invoice — and Balance activity shows your balance changes with usage summed per day. The money in your account is now called your balance everywhere: the setting is "At a limit, continue on my balance" and the key option is "Balance only" (the 402 plan_limit_reached message uses the new label; its status, type, code and Retry-After are unchanged). In the Playground each reply shows whether your Agent plan or your balance paid for it, and the sidebar's plan usage updates while you work.
v0.1.10 (2026-10-06): The Agent plan is now easy to find: it has its own item in the account sidebar, a panel above your balance with your tier and 5-hour usage (or the lowest monthly price when you have no plan), and a card on the account overview. Pricing now shows the two ways to pay — the Agent plan and pay as you go — as two tabs. The What is Tokener? and Quickstart guides describe both ways to pay.
v0.1.9 (2026-10-06): You can now change your Agent plan tier on the Plan page. An upgrade starts at once — Stripe charges only the prorated difference for the rest of the billing period, and the new limits apply immediately; a downgrade takes effect at the end of the current period and can be undone with "Keep" until then (how it works). "At a limit, continue on my credits" is now off by default for new subscriptions: turn it on to keep working on your credit balance when a limit is reached.
v0.1.8 (2026-10-06): New: the Agent plan, a monthly subscription that pays for your API and Playground usage up to a 5-hour and a weekly limit, on the same key and endpoints and every model. At a limit your requests continue on your credits if you allow it; otherwise they get a 402 plan_limit_reached with a Retry-After header. A key can be set to Credits only so it never uses the plan. The tiers are on Pricing, and the Agent plan guide explains the limits, the 402 response, Credits only keys and how to cancel.
v0.1.7 (2026-10-04): New models are on the Models page, and a model card may now show a precision tag (BF16 = unquantized weights, FP8 = an 8-bit release). A new Data handling setting lets an account also allow suppliers without a no-retention commitment, which can improve availability and is billed a percentage below list on every model (see Who processes my requests?). Reasoning levels and the default are now per model, listed on each model page (Reasoning controls). Image parts must be inline base64 data URLs — a remote image URL is rejected with a 400 (how to send an image). Token counting on the Anthropic-compatible (Messages) protocol is available only on accounts that allow all suppliers. A request body too large for its model now answers 413. Published performance figures on model pages are the median of the last 10 test requests.
v0.1.6 (2026-09-21): New guide: Claude Code — use the Claude Code CLI with Tokener's open models over the Anthropic-compatible (Messages) protocol, proven end to end before it was written. Linked from the Anthropic SDK guide and Developer Onboarding.
v0.1.5 (2026-09-20): The API reference now explains how usage numbers are counted and billed — input totals and their cached part on OpenAI-compatible (Chat Completions) and OpenAI Responses, why the Anthropic-compatible (Messages) protocol is billed from the input total, the charge as a formula, free token counting, and what the Usage page shows. The other guides link to it.
v0.1.4 (2026-09-20): Token counting is now available on the Anthropic-compatible (Messages) protocol — POST /v1/messages/count_tokens returns a request's input-token count and is free (see the Anthropic SDK guide). Messages usage is now billed from the request's input total.
v0.1.3 (2026-09-20): Price update: input and cached-input prices were lowered on one model — current prices are always on Pricing.
v0.1.2 (2026-09-19): New pages: Pricing (how pay-as-you-go billing works, with the live price list), FAQ, and About. A new API reference guide lists every endpoint, the accepted authentication headers, required request fields, streaming formats, usage fields and error codes for the three protocols. Every model now has its own page with prices, specs and a quick start.
v0.1.1 (2026-09-19): A "Start here" track for beginners: What is Tokener?, Quickstart without code and a Playground guide, plus a "New here?" box on the docs home.
v0.1.0 (2026-09-18): Three API protocols — OpenAI-compatible (Chat Completions), OpenAI Responses, and Anthropic-compatible (Messages) — over one metered account. The Playground gained a protocol selector with per-protocol Get Code, the Usage page now shows the protocol of every request, and two new guides landed: OpenAI Responses API and Anthropic SDK.
v0.0.1 (2026-09-01): First version released and live online.