Model / API profile · Checked September 27, 2026
Respan Span-01
Respan Span-01 is a reasoning classifier for agent and conversation traces. Give it a trace and natural-language behavior definitions; it returns classification probabilities rather than acting as a general chat assistant.
Release: 2026-09-24. Source-verified; not independently benchmarked.
What the classifications mean
The documented outputs distinguish present, absent and not-observable behavior. Several definitions can be scored for the same trace. This fits teams checking escalation rules, tool misuse, grounding or safety-related behavior. “Not observable” matters when the trace lacks enough evidence; absence of evidence should not silently become a clean pass.
API naming and access
Current Respan documentation uses POST /api/v1/scores, not a separately verified “Decisions API.” The documented model IDs are respan-01-free for Lite and respan-01-pro for the main model. Both remain subject to early-access organization enablement; the quickstart describes a 403 response before access is enabled.
Free and paid usage
Respan documents Lite as free with a daily cap resetting at 00:00 UTC. The main model is listed at $0.02 per million input tokens, counting the trace and definitions, with output free and charges using Respan credits. Free does not mean uncapped or immediately available to every account.
OpenRouter status and limitations
An OpenRouter Span-01 Lite listing was visible during research, but a read of its public model catalog returned no matching Span-01 entry. Routable OpenRouter availability was therefore not confirmed. Use Respan’s access instructions as the verified route and check your own account before planning an integration. No inference request was made.
Before relying on scores
Define observable behavior precisely and compare scores against a small, manually labeled set of permitted traces. Choose thresholds and human escalation rules for your task; probabilities are not a compliance certification or proof that an agent is safe. Do not send private traces until your team has approved that data flow. This profile reports documentation, not independent classifier accuracy.
Official sources
Checked September 27, 2026. Access and prices can change; follow the linked provider documentation before use.