Releasesby GlyphTraining Serverless Training deprecation: DeepSeek V4 Flash (0731) DeepSeek V4 Flash (0731) is removed from...
Training Serverless Training deprecation: DeepSeek V4 Flash (0731) DeepSeek V4 Flash (0731) is removed from the shared Serverless Training pool effective November 2, 2026. Migrate Serverless Training workloads to DeepSeek V4.1 Flash 1017 . This change applies only to Serverless Training.
Also shipped Nov 2, 2026Fireworks AI in November 2026 →
Managed warehouse for MLflow (Beta)
Managed warehouse for MLflow is available in Beta, enabling users to view Unity Catalog traces in the MLflow UI without manually choosing a SQL warehouse.
Check out GitHub code without long-lived credentials on self-hosted agents
Self-hosted agents can now use temporary, read-only GitHub App tokens for repository checkout. This eliminates the need to distribute and rotate long-lived SSH keys or personal access tokens across your agent fleet.
Serving tuned Gemini 3.x models on global endpoints
Serving tuned Gemini 3.x models on global endpoints is available in General Availability (GA) . You can copy fine-tuned Gemini 3.x models to the global region using the Model Registry models:copy API, deploy them to global serving endpoints, and send inference requests directly to aiplatform.googleapis.com .
[Launched] Generally Available: Change from virtual network integration to a private endpoint capable network configuration in Azure Database for PostgreSQL flexible server
You now have greater flexibility in network design for Azure Database for PostgreSQL Flexible Server workloads. Now you can change existing VNet-integrated servers from delegated-subnet networking to a private endpoint-capable configuration, enabling secu
[Launched] Generally Available: Restart databases in Azure SQL DB via the Azure portal
You can now restart or fail over databases directly from the Azure portal, including single databases, Hyperscale databases, and databases in elastic pools, without writing scripts against the Rest API or contacting support. Learn more.
Store and recover user sessions with RUM Managed Archive
Datadog RUM Managed Archive allows users to store all ingested RUM sessions and recover them on demand for debugging, incident investigation, and audits.
More from Fireworks AI platformEvery Fireworks AI platform release →
Inference New Serverless mode: Ultra-fast (Preview) Ultra-fast is a new high-speed Serverless mode, now in...
Inference New Serverless mode: Ultra-fast (Preview) Ultra-fast is a new high-speed Serverless mode, now in Preview, for highly latency-sensitive applications. Ultra-fast variants aim for 6-8x generation speed, and up to 10x for agentic coding tasks. Like Fast, it is not a different model, so output quality is the same.
Fireworks AI releases Training Cookbook on PyPI and migrates repository
Fireworks AI published fireworks-training-cookbook 2026.11.5.1 on PyPI for Python 3.11+ and migrated its training cookbook to fw-ai-external/cookbook.
Fireworks AI adds Rune 26B-A4B model to Serverless
Fireworks AI has added the Rune 26B-A4B model to its Serverless platform. This decision model processes documents and questions to return calibrated probabilities without generating text, and is billed based on input tokens only.
Errors identify their source
Fireworks AI updated its error reporting to provide more granular source identification for failures, including new error codes for parsing and transport issues.
Unparseable trajectories are discarded
Fireworks AI now discards unparseable trajectories to prevent them from impacting training data quality.
Failure traces are saved locally
Failed model trajectories now generate private JSONL traces in the system temporary directory. These traces include request data, token IDs, and exception chains to help developers identify whether failures originated from the model, tokenizer, or renderer.
Image inputs for FiretitanSamplingClient.sample
The sample and sample_async methods now support a keyword-only images argument. This allows developers to pass base64 data URLs for token prompts using placeholders.
Errors identify their source
Before 1.2.27, the sidecar returned HTTP 502 for two failures that weren't server-side: tito_model_malformed_output (unparseable model output) and tito_upstream_error (any sampling exception, including client timeouts and dropped connections). They are replaced by: tito_sdk_parse_error (422): the output couldn't be parsed into an assistant or tool message.