Field notes

notes fromthe subnet

How gm is built, what we learn running frontier models, and what verifiable inference actually takes.

FeaturedProduct AnnouncementsIntroducing SayGM: An AI Gateway Built Like a VaultEvery prompt you send to an AI model passes through someone else's servers first. SayGM is the AI gateway built so that "someone else" structurally can't read it — hardware-verified, not just policy-promised. Here's what that means, and why it's also cheaper than the alternatives.Aug 3, 20265 min read

Header image for an article outlining SayGm and it's AI gateway that's build like a vault.

EducationKimi K2.6 and K3 API Pricing: Every Provider ComparedKimi K2.6 costs $0.28 per million input tokens on SayGm and $0.95 direct from Moonshot. Here's what every major provider charges for K2.6 and K3, what 100 million tokens actually costs, and which host fits your workload.Sep 16, 20265 min
Case StudyOpenAI's Navier-Stokes Dispute: Who Can Read Your Prompts?OpenAI says nobody read the mathematicians' unpublished work before its Navier-Stokes result, and that data from their usage may have improved its models. Both can be true, and that is what separates access from training.Sep 9, 20265 min
EducationWhat Is Private Inference? A Guide to TEE-Based AIMost AI providers promise they won't read your prompts. Private inference replaces that promise with hardware that can't. Here's how a trusted execution environment works, what it costs, and how to tell one privacy claim from another before you commit.Sep 9, 20265 min
Case StudySayGm vs OpenRouter: Hardware Attestation vs. a Privacy PolicyIf you're evaluating openrouter alternatives, the usual sticking points are the platform fee, the model catalog, or a security question "we don't log your prompts by default" doesn't quite answer. Here's how SayGm's hardware-attested TEE actually stacks up against OpenRouter's policy-based privacy, model for model and dollar for dollar.Sep 4, 20265 min
Case StudyWhat is an LLM Gateway (and Why Attestation Matters)An LLM gateway used to be judged purely on routing and failover. What's changed is that whoever operates it is also structurally positioned to see everything passing through, which is why hardware attestation is starting to matter as much as uptime. Here's the difference between a gateway that promises it won't look and one that can't.Sep 4, 20265 min
Case StudyTop 5 OpenRouter Alternatives (With Real Pricing)OpenRouter earned its spot as the default multi-model gateway, but its 5.5% platform fee and policy-based privacy model have left room for real competition. This breakdown compares five alternatives, including SayGm's TEE-verified routing, with real, current pricing for each.Aug 28, 20265 min
EducationFixing "Provider Returned Error" on OpenRouter"Provider returned error" is a wrapper, not a diagnosis. Here's how to tell rate limits, provider outages, and malformed requests apart, and fix each one fast.Aug 20, 20265 min
EducationThe 5 Best LLMs to Use Right Now in 2026New frontier models ship every six weeks, so chasing every launch is a losing game. Here are the five actually worth using right now, and exactly what each one is best at.Aug 19, 20265 min
EducationBest Unified API Platforms: OpenRouter vs SayGm vs LiteLLMOpenRouter, LiteLLM, and SayGm all promise to collapse your model integrations into one API. Here's how they actually differ on cost, control, and what happens to your prompts along the way.Aug 19, 20265 min