Releasesby IbisTraining New Serverless Training model: Muse Spark Open Muse Spark Open is now available for LoRA workloads...
Update5 days agoUnverifiedAdded Oct 19, 2026
Training New Serverless Training model: Muse Spark Open Muse Spark Open is now available for LoRA workloads on the shared Serverless Training pool, with up to 262K context. Use accounts/fireworks/models/muse-spark-open-nvfp4 as the base model. See the Serverless Training guide for setup and the training model catalog for current availability.
Topics: Governance, Observability, Developer tools
More Fireworks AI platform releases
Every Fireworks AI platform releaseTraining Upcoming Serverless Training deprecation: DeepSeek V4 Flash (0731) On November 2, 2026 , DeepSeek V4...
Training Upcoming Serverless Training deprecation: DeepSeek V4 Flash (0731) On November 2, 2026 , DeepSeek V4 Flash (0731) will be removed from the shared Serverless Training pool. This change applies only to Serverless Training. If you train DeepSeek V4 Flash (0731) on Serverless Training, migrate to DeepSeek V4.1 Flash 1017 before November 2, 2026.
Action required
If you use any of the models below on serverless, migrate to a recommended replacement before November 7, 2026 . After that date, they will no longer be available via serverless endpoints.
Recommended migrations
- GPT OSS 120B , move to a dedicated deployment of GPT OSS 120B Inkling , migrate to Muse Spark Open (US endpoint, 50% US regional premium waived) NVIDIA Nemotron 3 Ultra , move to a dedicated deployment of Nemotron Super VL NVIDIA Nemotron 3.5 Lightning 30B A3B , move to a dedicated deployment of Nemotron Super VL MiniMax M3 (US endpoint only) , migrate to...
New models on Serverless
- DeepSeek V4.1 Flash 1017: accounts/fireworks/models/deepseek-v4p1-flash-1017 DeepSeek V4.1 Flash 1017 (US endpoint): accounts/fireworks/models/deepseek-v4p1-flash-1017-us Muse Spark Open (US endpoint, 50% US regional premium waived): accounts/fireworks/models/muse-spark-open MiniMax M3.1: coming to serverless after its open-source release, expected...
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.
Image inputs for FiretitanSamplingClient.sample
Since 1.2.27, sample and sample_async take a keyword-only images= list of base64 data URLs for token prompts with one placeholder per image. Prompts with unprojected image chunks are rejected so sampler-owned delimiters aren't duplicated. See the cookbook's TITO failure debugging reference for every error code and how to read a trace.
Also shipped on Oct 19, 2026
Fireworks AI in October 2026Cloud Workload Security is now available in Beta
Cloud Workload Security (CWS) is now available in Beta. Apply policy controls to cloud API requests before they reach your cloud provider. Define policies with request filters and Rego checks. For example, block requests that would make an R2 bucket public. Use workloads and policy sets to target your cloud environments.
Copilot model policy API for enterprise, orgs, and teams
You can now use the REST API to retrieve and update Copilot model policies for enterprises, enterprise teams, enterprise apps, and organizations. What’s new GET and PATCH endpoints are now generally available for model policies at four levels: Enterprise: /enterprises/{enterprise}/copilot/model-policies Enterprise teams...
GitHub Copilot models deprecated on October 19, 2026
As of today, October 19, 2026, we have deprecated the following models across most GitHub Copilot experiences (including Copilot Chat, inline edits, ask and agent modes, and code completions). These deprecations were announced on September 18, 2026 .
Postgres wait events, now in Query Insights
Query Insights for ClickHouse Managed Postgres now includes wait event breakdowns, blocked-session charts, and lock-wait charts. These metrics are also available in Prometheus.
Send OpenTelemetry traces to Agent Observability, no SDK required
You can now send OpenTelemetry traces straight to Agent Observability, without the SDK. If your agents already emit OpenTelemetry generative AI spans, from your framework, an instrumentation library, or spans you write yourself, keep your existing route through Grafana Alloy or an OpenTelemetry Collector and opt in your AI spans with a single attribute.
Cortex Tools for Cortex AI Gateway (Preview)
We are pleased to announce the preview of Cortex Tools, a part of Cortex AI Gateway that connects the AI clients your organization uses to the business tools your work already lives in, using the Model Context Protocol (MCP). Supported clients include CoCo Desktop, CoCo CLI, Claude, and Cursor.