SunsetDiff

Fireworks model shutdowns on November 7, 2026

The dataset has 4 Fireworks ids with shutdown date 2026-11-07: accounts/fireworks/models/gpt-oss-120b, accounts/fireworks/models/nemotron-3-ultra-nvfp4, accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b, accounts/fireworks/routers/glm-5p3-flash-us. The changelog says these serverless models are decommissioned on November 7, 2026, and that dedicated deployments and Reserved Throughput are unaffected. GPT OSS 120B and the two Nemotron rows have an empty replacement field. The changelog says to move those to a dedicated deployment and does not print a serverless model id. accounts/fireworks/routers/glm-5p3-flash-us stores accounts/fireworks/routers/deepseek-v4p1-flash-1017-us. The same changelog section also names Inkling and MiniMax M3 on the US endpoint. Those names are not in the table.

Last reviewed . Publisher: SunsetDiff.

Steps

  1. Read the four ids. The table is every Fireworks id in the dataset whose shutdown date is 2026-11-07. Not published means the stored replacement list is empty. Open the model page before you change a call.
  2. Separate serverless from dedicated. The changelog says this date applies only to serverless inference. A dedicated deployment or Reserved Throughput of the same model stays up. The three rows with an empty replacement say to move serverless traffic onto a dedicated deployment. They do not name a new model id.
  3. Change the US GLM id. Calls to accounts/fireworks/routers/glm-5p3-flash-us on https://us.api.fireworks.ai use accounts/fireworks/routers/deepseek-v4p1-flash-1017-us in the model field. The changelog says the global GLM 5.3 Flash endpoint remains available.
  4. Test before the date. Send one real prompt to the id you will call. Do that before November 7, 2026. A dedicated deployment is a different setup from a serverless model string.

Models shutting down

These rows are every Fireworks model in the dataset with shutdown date 2026-11-07. The model id links to its SunsetDiff model page. Deprecation date, shutdown date, and replacement are the fields stored on that row. Not published means the field is empty. These rows do not store a deprecation_date.

The changelog says: "Several Serverless models will be decommissioned on November 7, 2026 to better serve newer, higher-performance replacements. This applies only to serverless inference. Dedicated deployments and Reserved Throughput are unaffected." It also says: "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."

GPT OSS 120B says move to a dedicated deployment of GPT OSS 120B. Each Nemotron row says move to a dedicated deployment of Nemotron Super VL. Those sentences do not print a serverless model id, so the replacement cell is Not published. This guide does not fill one in.

accounts/fireworks/routers/glm-5p3-flash-us is the US endpoint id on the US-only serverless page. The stored replacement is accounts/fireworks/routers/deepseek-v4p1-flash-1017-us. The changelog says: "The global GLM 5.3 Flash endpoint remains available." It also says: "GLM 5.3 and DeepSeek V4.1 Flash are not affected and remain available on serverless."

4 Fireworks model ids with shutdown date 2026-11-07.
Model idDeprecation dateShutdown dateRecommended replacement
accounts/fireworks/models/gpt-oss-120bNot published2026-11-07Not published
accounts/fireworks/models/nemotron-3-ultra-nvfp4Not published2026-11-07Not published
accounts/fireworks/models/nemotron-lightning-3p5-30b-a3bNot published2026-11-07Not published
accounts/fireworks/routers/glm-5p3-flash-usNot published2026-11-07accounts/fireworks/routers/deepseek-v4p1-flash-1017-us

Sources: Fireworks changelog , Fireworks US-only serverless , checked November 1, 2026.

Names in the same section that are not stored

The changelog heading also names Inkling and the US endpoint for MiniMax M3. SunsetDiff stores a row when a deprecation links a model page, or when the US-only page prints the model id. A display name with no model id is not stored.

Inkling has no model link and no model id on that page. The changelog says to migrate to Muse Spark Open (US endpoint, 50% US regional premium waived). The new-models list on the same entry prints accounts/fireworks/models/muse-spark-open. This guide does not invent an Inkling id.

MiniMax M3 (US endpoint only) links the global model page. The changelog says to migrate to DeepSeek V4.1 Flash 1017 (US endpoint), and: "The global MiniMax M3 endpoint remains available." The US-only page does not print a MiniMax M3 US model id. The stored global row is accounts/fireworks/models/minimax-m3, status active, shutdown date Not published. The US id the changelog does print for that migration is accounts/fireworks/routers/deepseek-v4p1-flash-1017-us.

Sources: Fireworks changelog , Fireworks US-only serverless , checked November 1, 2026.

What breaks

On November 7, 2026 the four ids reach the shutdown date stored in the dataset. The changelog says that after that date they will no longer be available via serverless endpoints. It does not print an HTTP status.

The same entry says: "Several Serverless models will be decommissioned on November 7, 2026 to better serve newer, higher-performance replacements. This applies only to serverless inference. Dedicated deployments and Reserved Throughput are unaffected." A dedicated deployment that is already running is the case the changelog says is unaffected. A serverless call is the case that stops.

The US page says: "Call https://us.api.fireworks.ai and pass one of the US model IDs from the table above as model." The curl on that page posts to https://us.api.fireworks.ai/inference/v1/chat/completions. The quickstart's OpenAI-compatible host for a global id is https://api.fireworks.ai/inference/v1. accounts/fireworks/routers/glm-5p3-flash-us is a US id. Send it to the US host.

Sources: Fireworks changelog , Fireworks US-only serverless , Fireworks quickstart , checked November 1, 2026.

Find the model ids in your code

Search the repository for the four ids in the table. Also search for Inkling and for Muse Spark Open if a config uses the display name. A hit is a candidate, not proof the call still runs. How to find every model ID in your codebase covers SDK argument names. The scan page builds a search from dataset ids that have a shutdown date.

A US call and a global call can use different hosts. The US page's host is https://us.api.fireworks.ai. The quickstart's host is https://api.fireworks.ai.

ripgrep

rg -n -w -e 'accounts/fireworks/models/gpt-oss-120b' -e 'accounts/fireworks/models/nemotron-3-ultra-nvfp4' -e 'accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b' -e 'accounts/fireworks/routers/glm-5p3-flash-us'

Sources: Fireworks changelog , Fireworks US-only serverless , checked November 1, 2026.

Test the replacement before the cutoff

For the US GLM row, call accounts/fireworks/routers/deepseek-v4p1-flash-1017-us on the US host with the prompts you send now. Do that before November 7, 2026. The US page says: "Beginning September 1, 2026, launched US-only models are priced at 1.5x the base model serverless prices." Both accounts/fireworks/routers/glm-5p3-flash-us and accounts/fireworks/routers/deepseek-v4p1-flash-1017-us are in that page's model table. The changelog says the global GLM 5.3 Flash endpoint remains available, so a global call is a different endpoint.

For GPT OSS 120B and the two Nemotron rows, the changelog does not name a serverless model id. It points serverless users at a dedicated deployment and links the on-demand deployment quickstart. This guide does not invent the id to send.

The pre-migration test checklist is the longer list: evals, rejected fields, token limits, tools, price, and latency. These Fireworks rows do not store a price.

OpenAI-compatible client, US host, model field only

from openai import OpenAI
client = OpenAI(base_url="https://us.api.fireworks.ai/inference/v1")
response = client.chat.completions.create(
    model="accounts/fireworks/routers/deepseek-v4p1-flash-1017-us",
    messages=[{"role": "user", "content": "Hello"}],
)

Sources: Fireworks changelog , Fireworks US-only serverless , On-demand deployment quickstart , checked November 1, 2026.

Related SunsetDiff pages

All guides

Questions

Which Fireworks ids shut down on November 7, 2026?

The dataset lists accounts/fireworks/models/gpt-oss-120b, accounts/fireworks/models/nemotron-3-ultra-nvfp4, accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b, accounts/fireworks/routers/glm-5p3-flash-us with shutdown date 2026-11-07. The changelog says: "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."

What is the official replacement?

The recommended replacement is the official replacement stored on each row. accounts/fireworks/models/gpt-oss-120b: Not published. accounts/fireworks/models/nemotron-3-ultra-nvfp4: Not published. accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b: Not published. accounts/fireworks/routers/glm-5p3-flash-us: accounts/fireworks/routers/deepseek-v4p1-flash-1017-us. The changelog tells serverless users of GPT OSS 120B and the two Nemotron models to move to a dedicated deployment. It does not print a serverless model id for those three. accounts/fireworks/routers/glm-5p3-flash-us stores accounts/fireworks/routers/deepseek-v4p1-flash-1017-us.

Do dedicated deployments stop on that date?

The changelog says: "Several Serverless models will be decommissioned on November 7, 2026 to better serve newer, higher-performance replacements. This applies only to serverless inference. Dedicated deployments and Reserved Throughput are unaffected." Dedicated deployments and Reserved Throughput stay up. Serverless endpoints stop.

Why are Inkling and MiniMax M3 US missing from the table?

Inkling has no model id on the changelog. The page says to migrate it to Muse Spark Open and prints accounts/fireworks/models/muse-spark-open. MiniMax M3 (US endpoint only) links the global model. The US-only page does not print a MiniMax M3 US id. The changelog says: "The global MiniMax M3 endpoint remains available." The US id it does print for that migration is accounts/fireworks/routers/deepseek-v4p1-flash-1017-us.

Which host does the US GLM id use?

The US page says: "Call https://us.api.fireworks.ai and pass one of the US model IDs from the table above as model." Its curl posts to https://us.api.fireworks.ai/inference/v1/chat/completions. The quickstart host https://api.fireworks.ai/inference/v1 is the global host.

More guides

Deprecation alerts

We email you before models shut down. Unsubscribe anytime.

Deprecation alert signup
Providers (optional)