Union Alpha current-status, repeatability, and capacity delta
Run run:manual:8245aa52a11393edf3b6e8fbfc3ccabd · cutoff 2026-09-18T08:26:20Z–08:35:00Z UTC · narrow route/status recheck. This publication summarizes the accepted report; it preserves R1 separately.
unbiased/pareto; the public catalog has no old-ID match. OpenCode Go still documents union-alpha, but its subscription/key-gated access was not tested. R2 sent zero inference requests.Route-by-route result
| Route | Current official state | R2 observation | Classification / action |
|---|---|---|---|
OpenRouter stealth/union-alpha | Historical listing says Pareto by Unbiased; free stealth period ended. Live catalog has no old ID. | n=0. No attempt: old route no longer listed; no current credential. | No longer listed. Do not start a build on the old ID. Evaluate paid Pareto as a separate route. |
OpenCode Go union-alpha / opencode-go/union-alpha | Docs and public model list retain it; docs require Zen sign-in, Go subscription and key. | n=0. No pre-existing authorized key in the runner. | Documentation-only. An authorized subscriber can make one small non-sensitive request and capture its first HTTP result. |
Catalog omission does not reveal an old-ID HTTP response. A public Go model listing does not prove entitlement, successful inference or zero-cost access.
What changed from R1
| Field | R1 | R2 | Reader impact |
|---|---|---|---|
| OpenRouter route | Free stealth/union-alpha listing. | Preview ended; old ID absent; paid Pareto successor named. | Retire active/free preview instructions. |
| Attribution | Anonymous; no confirmed owner. | OpenRouter: Pareto by Unbiased. Unbiased: built/maintained by Circuit & Chisel. | Retire identity theories as the current answer; keep composition unknown. |
| Limits | No model-specific OpenRouter quota. | Still none for old ID; generic 50/day and colon-suffix policy remain generic. Go “Unlimited” remains limited-time documentation. | Do not call either a Union Alpha/Pareto quota or SLA. |
| Observations | Four DNS failures before HTTP. | No request because no authorized pre-existing credential. | R2 does not repeat or refute the historic DNS finding; capacity remains unmeasured. |
| Benchmark evidence | No qualifying public comparison. | Provider publishes Pareto 26.9 claims (74 DeepSWE, 51 TB4) plus a harness. | Provider claim only—not independent proof of the R1 preview snapshot. |
Limits, entitlement and observed results—separated
per_request_limits: null does not mean unlimited.Attribution and benchmark boundary
OpenRouter’s reveal and Unbiased’s own attribution resolve the former owner question. They do not reveal Pareto’s component membership or make prompt-fingerprint theories proof. The provider model card claims Pareto 26.9 scores of 74 DeepSWE and 51 Terminal-Bench 4.0; the public harness documents an intended method. The inspected materials did not expose raw runs linking those figures to the retired Union Alpha preview, so no verified Union Alpha chart is published.
Blockers and remaining unknowns
- R2 had no
OPENROUTER_API_KEYorOPENCODE_API_KEY; no account, subscription, key or purchase was created. - There is no R2 evidence of old-ID HTTP behavior, OpenCode authorization, routing, latency, tool/image support, reliability or capacity.
- Historical stealth retention wording must not be projected onto current Pareto without a Pareto-specific source. Go’s 0-day retention/no-training entry remains route-specific documentation.
- The relationship between the old preview, paid Pareto and Go alias remains unverified at a version/snapshot level.
Evidence register
Primary: historical OpenRouter listing (R2-S01), catalog (R2-S02), Pareto listing (R2-S03), limits (R2-S04), OpenCode Go docs (R2-S06), and Unbiased (R2-S09). Full source/probe artifacts are retained in the accepted run. Historical baseline: R1 launch audit.