Journey

Notes from building Metis in public — what we tried, what broke, and what worked.

Our Own Weights

Jul 7, 2026

Every entry so far ran on borrowed weights. Today a model trained from scratch on this MacBook — on our own code, for zero dollars — answered inside OpenCode. It's tiny, it's bad, and it changes what this project is.

Checking Is Cheap. Saying No Is Not.

Jun 26, 2026

We tried to push Metis's founding bet — that verifying is cheaper than generating — to its logical extreme, and measured it. The bet is real, but it splits in two: recognizing a true answer is free, rejecting a plausible false one is not, and the second half doesn't scale with model size.

How Small Can the Model Go?

Jun 25, 2026

If the architecture does the work, the model should be replaceable. We swapped the 1.7B generator for a 0.6B one, 62% less model, and measured what broke.

The Tie Broke (and We Found Something Better)

Jun 25, 2026

We built a harder benchmark to test whether 0.6B really equals 1.7B. It doesn't. But the bigger surprise was an architecture bug that was quietly costing the larger model nine right answers.

Teaching the Shortcut Some Restraint

Jun 25, 2026

The fast path was answering questions it had no business answering. The fix wasn't to remove it — it was to teach it which questions are actually shortcuts.

The Bet

Jun 24, 2026

Frontier models waste most of their parameters storing facts. What if you put the facts somewhere else?

The Physics Don't Lie

Jun 24, 2026

Before writing a line of code, you have to be honest about what fits in 4 GB and what doesn't.

Retrieval Collapses a 25x Parameter Gap

Jun 24, 2026

The most important finding in the research: a small model with retrieval routinely beats a model 25 times larger without it.

V1 Runs

Jun 24, 2026

The first version that actually works end to end: local inference, retrieval, verification, citation.

The 408-Second Answer

Jun 24, 2026

A single configuration variable was turning a 3-second answer into a 7-minute one. This is how we found it.

Checking Is Cheaper Than Generating

Jun 24, 2026

The research finding that makes the whole GVS architecture work: verifying a claim against evidence is fundamentally easier than producing a correct one from scratch.

The Benchmark

Jun 24, 2026

We ran qwen3:1.7b bare vs inside Metis on the same 14 questions. Same weights, same hardware, different architecture.

More entries coming as the build progresses.