Standard Intelligence

1. Organisations that require an off-the-shelf automation platform immediately If the objective is to automate AP/AR, HR workflows, or standardized back-office processes with established governance, mature RPA platforms are typically a lower-risk choice.

2. Organisations requiring public compliance artifacts (SOC 2 / ISO / HIPAA) for procurement SI does not publicly document these certifications.

3. Organisations that need predictable cost and immediate access SI has no public pricing and limited indications of a generally available API.

4. Organisations unable to tolerate agentic security risk Computer-control agents increase the blast radius of errors; if strict permissioning, approval flows, and sandboxing cannot be implemented, the approach is not recommended.

• SI does not publish a consolidated limitations list, roadmap, or GA availability schedule.
• Community discussion references undisclosed benchmarks; lack of a public benchmark suite is a limitation for buyers seeking objective comparisons. (https://news.ycombinator.com/item?id=47125014)

• Focus on video-native computer action modeling emphasizing compression and long-context feasibility. (Vendor claim) (https://si.inc/posts/fdm1/)
• Reported scale of internal dataset: SI reports an 11-million-hour screen recording dataset. (Vendor claim) (https://si.inc/posts/fdm1/)
• Detailed infrastructure descriptions: the 30PB storage cluster write-up provides cost and architectural details. (https://si.inc/posts/the-heap/)
• Open model release in speech (hertz-dev) with released checkpoints. (https://si.inc/posts/hertz-dev/)

SI does not appear to have a mature footprint on mainstream B2B review platforms:

• Gartner Peer Insights: No readily accessible product listing/reviews found for "Standard Intelligence" (searches returned general Gartner pages rather than SI reviews).
• G2: No clear SI product profile/review page identified via accessible searches.
• TrustRadius / Capterra: No clear SI product profile/reviews identified via accessible searches.

Interpretation: This is consistent with SI operating primarily as a research lab without a broadly deployed commercial SaaS offering.

Because classic customer reviews are scarce, the most actionable user sentiment is from technical community discussion.

Hacker News (discussion around FDM-1):

• An SI team member introduced the post and provided details on dataset scale and model behavior; community members discussed data sourcing and access questions. (Hacker News discussion thread)
• On data sourcing: a commenter asked if data was synthesized; SI replied that the data was collected rather than synthesized. (Hacker News discussion thread)
• Access/productization questions included requests for information on Hugging Face or an API. (Hacker News discussion thread)
• SI response suggested future commercialization options such as API access or licensing. (Hacker News discussion thread)

LinkedIn amplification (not customer reviews): posts summarize SI’s claims about large-scale compute and compression efficiency and provide commentary; these are opinionated summaries rather than verified buyer outcomes. (LinkedIn posts)

SI’s public case studies are primarily demonstrations:

• CAD workflows (e.g., Blender gear extrusion sequence). (Vendor demo post: CAD workflow)
• A self-driving fine-tune demonstration using openpilot in "joystick mode," reporting a brief amount of fine-tuning data and driving around a block in San Francisco. (Vendor demo post: autonomous driving)
• Automated UI testing/fuzzing demo finding a bug in a toy banking app. (Vendor demo post: automated UI testing)

Critical evaluation:

• These are vendor-created demonstrations with limited methodological detail about failure cases, success rates, and safety constraints.
• Demonstrations do not establish production readiness: long-tail UI variance, authentication, rate limits, and enterprise guardrails are not validated in public materials.

• Vendor blog post describes building a multi-petabyte cluster and provides itemized costs and recurring expenses. (Vendor blog post: infrastructure cost breakdown)

Critical evaluation:

• The cost breakdown contains practical operational details useful for teams building similar infrastructure.
• The described configuration reflects trade-offs appropriate for research workloads (e.g., lower redundancy) that may not meet enterprise data system requirements. (Vendor blog post: infrastructure cost breakdown)

• Org type: AI-focused product companies, applied AI research teams, security research groups, and automation platform teams.
• Size: Likely mid-market to enterprise R&D groups that can fund experimentation; SI itself is a small organization.
• Roles: Head of AI/ML, applied research lead, automation platform lead, QA/testing engineering leadership, security engineering leadership.
• Pain points:
  • Need for agents that can handle long-horizon tasks spanning extended interactive sessions in GUI environments
  • Brittleness of RPA scripts due to UI drift
  • Need for deeper exploration-based testing (state-space exploration)

Standard Intelligence (SI) is a small AI research lab/public-benefit corporation ("standard intelligence pbc.") focused on building aligned AGI and, more concretely today, foundation models that can operate computers directly from video (its "forward dynamics model" series, including FDM-1) and an open conversational speech base model (hertz-dev). SI’s public materials present the organization as an applied research lab rather than a conventional SaaS vendor: it publishes technical blog posts, releases research artifacts (e.g., model checkpoints for hertz-dev), and indicates a future commercialization path via API/licensing rather than an off-the-shelf enterprise product.

Key publicly described capabilities:

• FDM-1: a computer-action model trained on a large screen-recording dataset; demos include long-horizon GUI workflows (e.g., CAD sequences), website exploration/fuzzing, and a self-driving fine-tune demonstration. (Vendor claims) (Vendor blog post: FDM-1)
• hertz-dev: a mid-sized full-duplex conversational speech base model released as open weights/checkpoints. (Vendor blog post: hertz-dev)
• Data infrastructure emphasis: SI built "the heap," described as a very large storage cluster and a core enabler for video-scale pretraining. (Vendor blog post: the-heap)

Current positioning (inferred from public materials):

• AI research organization / frontier-model builder, describing aims related to aligned general learners. (Vendor blog and public materials)
• Likely near-term buyers/partners: teams interested in computer-use agents (desktop/web automation), UI testing/fuzzing, and potentially specialized agentic workflows (CAD/engineering) if SI offers API access.

Caveat: SI does not publish a typical enterprise product page, feature matrix, SLAs, or customer list—so target market is primarily inferred from blog posts and third-party coverage.