--- title: "Fireworks AI API changes, deprecations and deadlines | Harbyn" description: "Fireworks AI: 1 upcoming deadline and 18 recorded changes as of 2026-10-25, each with its source." canonical: "https://harbyn.com/vendors/fireworks" last-updated: 2026-10-25 --- # Fireworks AI: API changes and deadlines Markdown version of . AI models and APIs. 1 upcoming deadline, 18 recorded changes. Every entry cites the Fireworks AI page we read it from. ## What we read The Fireworks AI pages and feeds our ingestion visits, politely and on a schedule. - **Changelog**: - **Changelog**: ## What the scanner looks for How `harbyn scan` recognises Fireworks AI in a repository, on your machine. Nothing leaves it. - **Packages**: `pypi fireworks-ai` - **API hosts**: `api.fireworks.ai` - **Model IDs starting with**: `accounts/fireworks/models/` ## Upcoming deadlines | Deadline | Change | Use instead | Details | Source | | --- | --- | --- | --- | --- | | Nov 2, 2026 | DeepSeek V4 Flash 0731 retires from Serverless Training November 2 | - | | | ## Change history Newest first, by the date Fireworks AI announced it. This is the archive: past changes are not warnings, and changes we recorded after they took effect are marked as such. ### October 2026 - Oct 18, 2026, Rate limit: [Expanded Small serverless rate-limit tier](https://docs.fireworks.ai/updates/changelog.md) - Oct 18, 2026, Feature: [DeepSeek V4.1 Flash 1017 available for Serverless Training](https://docs.fireworks.ai/updates/changelog.md) - Oct 18, 2026, Feature: [NVIDIA Nemotron 3.5 Super VL 120B A12B available for training](https://docs.fireworks.ai/updates/changelog.md) - Oct 15, 2026, Feature: [Muse Spark Open available for Serverless Training](https://docs.fireworks.ai/updates/changelog.md) - Oct 7, 2026, Rate limit: [Doubled serverless rate limits for Small models](https://docs.fireworks.ai/updates/changelog.md) - Oct 2, 2026, Rate limit: [Higher serverless rate limits for Small models](https://docs.fireworks.ai/updates/changelog.md) - Oct 1, 2026, Breaking change: [Managed RFT paused, no new reinforcement fine-tuning jobs accepted](https://harbyn.com/changes/fireworks/2026-10-01-managed-rft-is-paused-try-rl-on-the-training-api.md) - Oct 1, 2026, Policy: [Serverless pricing update for DeepSeek V4.1 Flash](https://docs.fireworks.ai/updates/changelog.md) (effective Oct 1, 2026; recorded after the fact) ### September 2026 - Sep 28, 2026, Feature: [DeepSeek V4.1 Flash available for LoRA training](https://docs.fireworks.ai/updates/changelog.md) - Sep 22, 2026, Feature: [GLM 5.3 Flash available for Serverless Training](https://docs.fireworks.ai/updates/changelog.md) - Sep 16, 2026, Deprecation: [Creating a deployment without a shape will soon require acceptShapelessRisk](https://harbyn.com/changes/fireworks/2026-09-16-new-deployment-creation-flags-deploymentshape-default-and-acceptshapelessrisk.md) - Sep 16, 2026, Feature: [GLM 5.3 available for Serverless Training](https://docs.fireworks.ai/updates/changelog.md) - Sep 13, 2026, Feature: [GLM 5.3 Flash available for LoRA training](https://docs.fireworks.ai/updates/changelog.md) - Sep 12, 2026, Retirement: [Older DeepSeek, GLM, Muse and Kimi models removed from Serverless on September 25, 2026](https://harbyn.com/changes/fireworks/2026-09-12-upcoming-serverless-deprecation-older-deepseek-glm-muse-and-kimi-models.md) (effective Sep 25, 2026; recorded after the fact) - Sep 8, 2026, Breaking change: [Deployment annotation keys must start with custom/ in the REST API](https://harbyn.com/changes/fireworks/2026-09-08-deployment-tags-and-annotation-api-changes.md) - Sep 7, 2026, Feature: [GLM 5.3 available for LoRA training on the Dedicated Training API](https://docs.fireworks.ai/updates/changelog.md) - Sep 1, 2026, Rate limit: [Serverless rate limit ceilings now scale by model size](https://docs.fireworks.ai/updates/changelog.md)