JC·HARNESS

Models

Gateway readiness, runtime configuration, and source-backed model details for Jean-Claude Hunt, Validate, Gapfill, and Score stages.

New Run
Models
3Canonical Jean-Claude model slugs.
Gateway Ready
3 / 3Based on AI Gateway server/runtime key presence.
Hunt Default
fable5Primary class execution route.
Score Default
opus48Comparative scoring route.

Configuration Boundary

This page can add one-action browser runtime keys through sessionStorage, but persistent deployment configuration still belongs in Vercel, Docker Compose, or the host environment. Raw key values are never rendered back by the API.
Model credential controlssession overrides

Runtime Credentials

Server credentials are used for downstream calls. A browser-session PlatPhorm key authenticates explicit operator actions when no Authorization header is present.

0 session keys
PlatPhorm API keyserver configured

Downstream PlatPhorm credential is present server-side; browser operator auth is still explicit.

Vercel AI Gateway keyserver configured

Used for model routing checks, future gateway-backed model calls.

Vercel tokenserver configured

Used for future sandbox job creation when server OIDC is unavailable.

Direct provider keys are hidden because AI Gateway is configured server-side for model execution.
Stage Routing
Default stage-to-model assignments used by web planning and the Jean-Claude operator flow.
hunt
Fable 5
anthropic/claude-fable-5
validate
Fable 5
anthropic/claude-fable-5
score
Claude Opus 4.8
anthropic/claude-opus-4.8
gapfill
Claude Opus 4.8
anthropic/claude-opus-4.8

Fable 5fable5

Primary Hunt + Validate

gateway configured unverified model
Gateway model
anthropic/claude-fable-5
Harness model id
claude-fable-5
Provider
anthropic
Key source
vercel-ai-gateway
Validate pin
yes
Fallback models
openai/gpt-5.4-nano

Pricing Inputs

Input
$3/M tokens
Output
$15/M tokens

Pricing is used for budget estimates and cost-event summaries. Final spend comes from persisted execution metadata, not a static page claim.

Source and Harness Notes

Anthropic Mythos-class frontier model for general hunt + validate use. Known refusal classes (operator trust required): auth-bypass, CORS, SSRF, OIDC-logic, injection, agent-impersonation.

Provider documentation

Claude Opus 4.8opus48

Scorer of record + gapfill

gateway configured unverified model
Gateway model
anthropic/claude-opus-4.8
Harness model id
claude-opus-4-8
Provider
anthropic
Key source
vercel-ai-gateway
Validate pin
no
Fallback models
openai/gpt-5.4-nano

Pricing Inputs

Input
$15/M tokens
Output
$75/M tokens

Pricing is used for budget estimates and cost-event summaries. Final spend comes from persisted execution metadata, not a static page claim.

Source and Harness Notes

Scorer of record for the comparative scorecard (spec § 6). Recommended for auth/SSRF/OIDC/CORS logic tasks when Fable 5 is blocked (dual-model pattern).

Provider documentation

GPT-5.5gpt55

Cross-validation

gateway configured unverified model
Gateway model
openai/gpt-5.5
Harness model id
gpt-5.5
Provider
openai
Key source
vercel-ai-gateway
Validate pin
no
Fallback models
openai/gpt-5.4-nano

Pricing Inputs

Input
$7/M tokens
Output
$28/M tokens

Pricing is used for budget estimates and cost-event summaries. Final spend comes from persisted execution metadata, not a static page claim.

Source and Harness Notes

OpenAI frontier model for cross-model validation and scoring diversity.

Provider documentation