Fireworks AI
fireworks.ai · Compute · 1 record · JSON
Fireworks AI, Muse Compute
trialemailno cardDretention-clamped
| Free tier | A free first LoRA fine-tune with no card: 'Muse Glimmer 30B SFT is free via UI and Serverless Training API until 11/30. No credit card required.' The promotion applies to LoRA SFT jobs started before 2026-12-01T00:00:00-08:00 and covers training only: 'RL, DPO, full-parameter training, deployments, and inference outside the training run are not covered.' In-session evaluation sampling is billed at the published per-token rates. unit none; metered on wall-clock; promotional-window |
|---|---|
| Price | none stated |
| Per-object cap | not applicable |
| Retention | the 2026-11-30 window binds, not a quota: there is no published cap on the number of free LoRA SFT jobs inside the window, and the day the window closes the same job is refused unless it is run with --allow-paid. |
| Egress | not stated |
| Identity | email no card on the free path: the cookbook states 'Muse Glimmer 30B SFT is free via UI and Serverless Training API until 11/30. No credit card required.' A payment method is added only 'to buy credits' on the general platform, and no hold is stated. The signup form is not quoted, so the email requirement is inferred rather than read; debit and prepaid acceptance are not stated. |
| Limits |
the 2026-11-30 window binds, not a quota: there is no published cap on the number of free LoRA SFT jobs inside the window, and the day the window closes the same job is refused unless it is run with --allow-paid. |
| Binding | retention-clamped the 2026-11-30 window binds, not a quota: there is no published cap on the number of free LoRA SFT jobs inside the window, and the day the window closes the same job is refused unless it is run with --allow-paid. |
| Interface | rest inferred from the tooling the provider documents; not probed |
| Tooling | not stated |
| Composes with | this is the only lane in this class that touches training, and it composes with the free inference lanes: train the adapter here by 2026-11-30, then serve the tuned model on a free recurrent lane, because deployments on Fireworks itself are not covered. |
| Reliability | no reliability event recorded |
| Notes | Important boundary for this assignment: a token allowance cannot train a model, and this is the exception that proves it, because what is free here is the training loop and not the tokens. Two limits are open: no cap on the number of jobs is published, and the export path for the adapter is not stated, so whether the tuned weights can leave the platform is unverified. Do not read this as a general free-training lane: the promotion names one 30B base model and four excluded job types. |
| Evidence rung | documented |
| Checked | 2026-10-22 (live); stale on 2026-11-21 |
Evidence (4)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | docs.fireworks.ai/fine-tuning/training-api/cookbook/muse-gli | 'Muse Glimmer 30B SFT is free via UI and Serverless Training API until 11/30. No credit card required.', quoted |
| 2026-10-22 | docs.fireworks.ai/fine-tuning/training-api/cookbook/muse-gli | 'applies the promotion to LoRA SFT jobs started before 2026-12-01T00:00:00-08:00 and refuses to launch after that unless you pass --allow-paid. RL, DPO, full-parameter training, deployments, and inference outside the training run are not covered', quoted |
| 2026-10-22 | docs.fireworks.ai/fine-tuning/training-api/cookbook/muse-gli | 'The promotion covers training. In-session evaluation sampling is priced separately at the published per-token rates; the default run samples 80 answers.', quoted |
| 2026-10-22 | fireworks.ai/pricing | 'Simply add a payment method to buy credits' for the general platform; the free credit line is gone from the pricing page, quoted |
More fields (1)
{
"wave2": {
"class": "promotional-window",
"free_tier_kind": "promotional-window",
"shape": {
"vcpu": null,
"ram_gib": null,
"gpu_class": null,
"gpu_count": 1
}
}
}Unknown or withheld values (7)
$.egress.meteredegress-unknown$.egress.termsnot-stated$.per_object_cap.notesprocess$.per_object_cap.statednot-applicable$.reliability_notenone-observed$.retention.notesprocess$.toolingnot-stated