Step 5 Preview
Flagship preview for long-horizon agentic work, software engineering, and professional tasks; the official card gives deployment configurations requiring 4 B300 or 8 H200 GPUs for FP8 examples.
Score calculation
(2 + 2 + 5 + 1 + 7) ÷ 5 = 3.4
5 of six dimensions rated. Unrated dimensions are excluded from the average.
Full scoring method ↗Setup details
StepFun Step 5 Preview, a publicly downloadable but explicitly preview-stage 600B/27B-active multimodal checkpoint. It is not treated as a stable general release.
What fits your needs
StepFun Step 5 Preview, a publicly downloadable but explicitly preview-stage 600B/27B-active multimodal checkpoint. It is not treated as a stable general release.
Good fit
A downloadable 600B total / 27B active multimodal model for high-end agentic work. The official card calls this a preview.
Tradeoffs to expect
Tradeoffs
StepFun lists the repository license as “other,” and serving examples require multiple high-end accelerators. Treat preview status and license as material constraints.
How to switch
Local setup
The official card gives FP8 examples with 4 B300 or 8 H200 GPUs, and BF16 examples with 8 B300 or 16 H200 GPUs. Check exact memory and the license files before deployment.
Practical questions
Weights and license
Weights are published, but the card lists license as “other”; read LICENSE.txt and LICENSE-CN.txt for the exact use terms.
Hardware
Official deployment examples use multiple B300 or H200 GPUs. This is a server-class setup, not an assumed consumer-laptop fit.
Training data
No complete named-source training data or source-level permission ledger is established in the reviewed preview card.
This is a practices and disclosure grade, not an estimate of energy or emissions for inference.
See the evidence
The model card gives GPU-count examples for deployment, which defines a runtime boundary. It provides no measured electricity, emissions, or water, so no impact estimate is supported.
1 + 1 reporting + 0 electricity + 0 controls = 2/10. This grades documented practices and transparency.
No training or per-request electricity, emissions, water, power sourcing, or independently reviewed footprint is established.
- Reporting1/3
- The official release materials provide downloadable weights and a local serving path for Step 5 Preview; this establishes an operating boundary, not a measured footprint.
- Electricity sourcing0/3
- No model-specific electricity sourcing or hourly power evidence is published in the reviewed sources; zero means unestablished, not dirty electricity.
- Resource controls0/3
- The card provides deployment configurations but no measured physical resource reductions or model-specific demand-control evidence.
Water disclosure
Water reporting has not been assessed for this choice.
The preview release publishes architecture and serving detail but no named-source training ledger or permission record.
See the evidence
The official Step 5 Preview materials focus on agentic capabilities, checkpoint details, and serving configurations. They do not identify the full training data or provide source-level permission evidence; a release license governs use of weights, not training-data rights.
1 base + 1 source transparency + 0 documented training data permission + 0 creator controls = 2/10; weight licensing is not treated as training-data permission.
Named dataset inventory, inherited base sources, rights for copyrighted or human-authored training material, and creator removal or retraining mechanisms remain unestablished.
- Source transparency1/3
- The model card describes architecture and checkpoint variants but does not provide a named training dataset manifest.
- Documented permission0/4
- The model card and license files don’t settle permission for the full training data.
- Creator controls0/2
- No source-level creator opt-out, removal, or retraining remedy is documented in the reviewed release materials.
Privacy depends on the hosting route; this preview checkpoint does not define a universal prompt-handling policy.
See the evidence
The official card provides self-hosted inference configurations and public weights, but it does not set one inference host or data-retention policy for every user. Local and hosted deployments have different privacy controls, so a universal model-level privacy score would mislead.
The selected host, region, telemetry, logging, and provider training terms are separate from the checkpoint release and require route-specific review.
A downloadable preview checkpoint and model card are public, while model-specific license terms and preview status limit openness and stability.
See the evidence
StepFun publishes model weights, inference code, and a detailed deployment guide, but the repository identifies the license as “other” and provides separate license files. The card labels the release Preview; review those terms and release status before reuse.
The weights and technical details are published, but the “other” license and preview-only status leave important use and stability limits.
The implications of the model-specific license for a particular deployment, preview-support timeline, and a full reproducible training recipe are not established here.
No worker protections or supply-chain labor assurance are established by the Step 5 Preview release materials.
See the evidence
The model page describes agentic capabilities, hardware, and deployment, not conditions for employees, data annotators, evaluators, moderators, or infrastructure contractors. This is a source-bounded disclosure gap, not evidence of mistreatment.
1 base + 0 conditions + 0 worker voice/remedy + 0 evidence/follow-through = 1/10; this is a disclosure finding, not proof of abuse.
Worker locations, pay, supplier coverage, worker voice, labor audits, and remediation outcomes are not established by the cited model-release sources.
- Pay & conditions0/3
- No worker-level conditions or supply-chain safeguards are established in this model-family source set. This is a disclosure gap, not proof of mistreatment.
- Worker voice & remedy0/3
- The model sources do not document a worker voice or remedy mechanism for data, evaluation, or infrastructure labor.
- Evidence & follow-through0/3
- No independent worker-level audit or remediation outcome is established in the reviewed model sources.
No military relationship or release-specific use safeguards are established for Step 5 Preview in the reviewed sources.
See the evidence
The official Step 5 Preview card identifies capabilities, release files, and serving configurations but does not report a military customer, contract, deployment, or military-use safeguards. A relationship may still exist even if this review found no record of it.
7 base + 0 safeguard adjustment = 7/10 (0 of 9 safeguard points; every 3 adds 1). No military relationship was established in the cited sources. This ranks above known defense ties without claiming a verified clean record.
Undisclosed customer contracts, deployment, end-use restrictions, and independent enforcement outcomes are not established by this release review.
- Use restrictions0/4
- The cited model license and release documents don’t show a model-specific binding ban on military or intelligence use. This is limited to this source set.
- Enforcement & oversight0/3
- No military-specific monitoring, enforcement, or remedy mechanism is established in the reviewed model materials.
- Relationship disclosure0/2
- The model materials do not disclose a model-specific military relationship or military-use record. A relationship may still exist even if this review found no record of it.
The reviewed model-release sources don’t confirm a defense relationship; absence of disclosure is not a clean-record finding.
Military rubric ↗StepFun is not named in the two official BNC campaign lists checked. This is a limited list check, not BDS approval.
See the evidence
The BNC consumer boycott/pressure guide and No Tech campaign page do not name StepFun among their targets. BDS uses selective campaigns, so absence from these two lists is neutral and should not be read as a clean record or approval.
This check covers only the named targets in the two cited BNC campaigns, not every BDS campaign, local group, investment, supplier, or downstream host. Lists and ownership may change.