Releasesby GlyphInference 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.
Also shipped Nov 3, 2026Fireworks AI in November 2026 →
Gemini Enterprise: Salesforce federated data store
Connecting a Salesforce data source with Gemini Enterprise using data federation is generally available (GA). You can use the Gemini Enterprise app assistant to search across your Salesforce accounts, contacts, opportunities, cases, leads, and custom objects, get answers grounded in that content, and create, update, and delete records without needing to...
[In preview] Public Preview: Extend immutability for PITR backups for Azure SQL Database and Azure SQL Managed Instance
With this enhancement, you have the ability to lock your retention to the configured PITR retention, beyond the default 7 days of immutable backups that are already available. This is especially useful for financial customers who typically have 35 days of
UI improvements now generally available for PD Reliability Platform customers
Customers on PD Reliability Platform now have an improved UI experience, including: Consolidated settings: Brings all your configuration and admin settings together in one place. New nav: Provides streamlined access to your most important capabilities. Your Subscription: Shows where the opportunities are to shift work off humans and onto AI and automation.
Claude auto-connect (beta) for PagerDuty MCP Server now in Early Access
With Claude auto-connect, members with a matching PagerDuty account are connected to PagerDuty in Claude automatically, with no sign-in screen and no identity provider (IdP) configuration. Members act as themselves in PagerDuty, with the permissions they already have there.
In BigQuery Studio , the left pane has been updated with new tab names
In BigQuery Studio , the left pane has been updated with new tab names. The Classic Explorer tab has been renamed to the Data tab, and the Repository tab has been renamed to the Source control tab. The Data tab now includes the Add data button and focuses on data hierarchies. These changes are generally available (GA).
Dependabot version updates support Agent Package Manager
You can now keep your AI agent skills, prompts, and instructions up to date with Dependabot version updates for Agent Package Manager (APM) . Dependabot opens pull requests to update supported git dependencies declared in apm.yml and regenerates apm.lock.yaml when it’s present. This support covers version updates, not security updates.
More from Fireworks AI platformEvery Fireworks AI platform release →
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.
Unparseable trajectories are discarded
A parse failure makes the whole trajectory untrainable, including earlier successful turns. The cookbook RL loop drops it without retrying or stopping the run, and other rollouts continue. Parse failures don't count toward the infrastructure circuit breaker, which now classifies transient failures more accurately and reports failure counts by reason.
Failure traces are saved locally
Each failed trajectory leaves a private JSONL trace (request, prepared token IDs, raw completion, exception chain) in fireworks-tito-traces under the system temp directory, capped at 256 MiB. Traces from clean trajectories are deleted, and failures rejected by cookbook adapters use the same location and limits.