Product updates and changes
Follow changes to model access, pricing, supported features, and integration requirements.
· Legal
Data policy is now part of the Terms of service.
Our legal documents are now the Terms of service and Privacy policy. Data processing, retention, model training, restricted data, and security provisions have moved into the Terms with their existing conditions preserved. Existing data-policy and security links lead to the Terms. This consolidation does not change retention periods or the conditions for earlier content or billing periods.
· Coding Plan
Coding Plan describes usage in plain terms.
The Coding Plan page now states its conditions in plain terms: no 5-hour window, no weekly quota, no monthly token allowance, and a baseline kept at peak hours. Prices and the model lineup are unchanged.
· Standard One
Standard One input rates updated.
One 3B now costs $0.018 per 1M text input tokens and $0.055 per 1M image input tokens, against list rates of $0.038 and $0.10. One 8B now costs $0.035 for text and $0.08 for image input, against list rates of $0.05 and $0.12. Output remains free.
· Legal
Zero Data Retention now covers Standard API, Standard One, and Product Plan.
The Terms, Privacy Policy and Data Processing Policy name Standard API (including Standard One) and Product Plan inference as Zero Data Retention services, and say that the Product Plan data platform is not yet released. Service metadata and restricted data are now defined once, in the Data Processing Policy. The Coding Plan terms state the billing periods: 7 days for Weekly and 30 days for Monthly. The Terms say that model availability depends on capacity and that a listed model may be unavailable or not yet released. The service level agreement says it does not apply to Coding Plan.
· Product Plan
Product Plan units now include a higher base allocation.
One unit’s guaranteed base allocation of aggregate output tokens per second is now 80 for Kimi K3, 135 for GLM-5.3, and 550 for DeepSeek V4.1 Flash, up from 40, 100, and 300. The price is unchanged: $1,200 per unit, per model, per month, and $3,000 for the three-model bundle. The Product Plan capacity calculator and the Resources guides that quote the rates now use the new figures.
· Coding Plan
Coding Plan is now one plan, weekly or monthly.
Weekly is $149 per developer, per week, and Monthly is $549 per developer, per month. Both include Kimi K3, GLM-5.3, and DeepSeek V4.1 Flash. No usage cap. Three requests at once at a guaranteed base rate of 80 tokens/s each, even at peak. Up to eight at once when capacity allows. Coding Plan may serve an application’s end users under the same conditions as Standard API. The new Coding Plan terms state billing, the capacity guarantee, the service credit, and model training.
Update, October 12, 2026: current usage conditions are on the Coding Plan page and Terms §20.3.
· Standard One
Standard One arrives in two open-weight sizes.
Standard One 3B and Standard One 8B read a message, an image or both, and answer each question with one of your options and a probability. Both are open weights under Apache 2.0 on Hugging Face. At launch, rates were $0.019 and $0.049 per 1M text input tokens, 50% off list, with output not billed. Input rates were updated on October 7, 2026. The accuracy, latency and calibration results in the write-up are measured on the served endpoint.
· API
GLM-5.3-Flash joins the API model list.
GLM-5.3-Flash takes text, image and video input with a 1.05M-token context, at $0.1275 per 1M input tokens and $0.425 per 1M output tokens. GLM-5.2, Kimi-K2.6 and Kimi-K2.7-Code are no longer listed.
· Resources
Two guides for deciding with Standard One.
The Resources library adds a guide to setting the confidence threshold at which an LLM’s answer ships without review, and a comparison of when a fine-tuned classifier beats an LLM on accuracy, speed and cost. Standard One’s figures come from its model cards, and every third-party figure is quoted from its source with the date it was read.
· Resources
A comparison of how coding plans count usage.
The Resources library adds a comparison of six coding subscriptions: what one unit of allowance counts at each vendor, whether the formula behind it is published, and when each 5-hour, weekly or monthly window refills. Every third-party figure is quoted from the vendor’s own page with the date it was read.
· Resources
Two references for comparing reserved capacity.
The Resources library adds a comparison of how nine providers sell reserved capacity, covering the unit each one measures, the billing clock, the minimum term, and whether the price is published, and a glossary of the capacity and plan terms those offerings use. Every third-party figure is quoted from the vendor’s own page with the date it was read.
· Resources
Three resources for sizing a Product Plan.
The Resources library adds a Product Plan capacity and break-even calculator, a guide to buying AI inference for a product on a monthly plan, and a guide to what an aggregate output rate means for a product. Every third-party figure is quoted from the vendor’s own page with the date it was read. Further comparisons and guides follow over the coming weeks.
· Site
Product Plan and Coding Plan have new addresses.
The two plan pages now live at /product-plan/ and /coding-plan/, and the pricing page’s sections and contact topics use the same names. The previous addresses redirect permanently, including links with query strings.
· API
Model cards show input modalities and rates.
Each card in the model catalog now names the input types the model accepts and its Standard API rates, so a model can be compared without leaving the page.
· Legal
Terms and data processing policy revised.
Zero Data Retention is stated to apply only where expressly identified for a service or configuration, optional storage through the Product Plan data platform is described as customer-directed, and Section 5.3 of the Terms now sets the conditions under which Service-Specific Terms may permit model training: the covered service and content, the purposes, the retention period, and the controls must be named before that use applies.
· Pricing
Standard API rates show each developer’s list price.
The rate table shows each model developer’s published list price beside our rate. DeepSeek-V4.1-Flash joined the catalog and the retired DeepSeek-V4-Flash-0731 was removed.
· API documentation
API contract and deployment terms clarified.
The site now describes the API contract and clarifies that regional processing controls apply when included in an order.