EVENT · E
Fable 5 is unavailable. What now?
Not a model review — a decision graph for routing your tasks after a capability disruption.
OCG LabOCG BenchFable 5 Unavailable
Live
TL;DR
Small or local coding tasks:use Opus 4.8
Large repo migration:decompose into bounded sub-tasks
Security-sensitive work:add deterministic and human validation
Cost-sensitive batch work:route routine chunks to Sonnet 4.6
Decision Card
Single-file bug fixOpus 4.8 or Sonnet 4.6
Multi-file refactorOpus 4.8 + test harness
Autonomous large migrationsplit / queue / validate
Security reviewmulti-model + human gate
High-volume routine transformationsSonnet 4.6
Terminal-heavy debuggingGPT-5.5 if CLI integration matters
Cognitive Event Graph
E changes the availability of r@D. The graph maps lost, degraded, fallback, substitute, decompose, and validate relations into C'.
EFable 5 unavailable
lost@E,Drepair + locate weakenedrepair + locate
degraded@E,Dlong-horizon coherence reducedgeneralize + track_state
fallback@Drestricted domains route downroute + scope
substitute@taskmap task to available modelselect + route
decompose@tasksplit when no direct substitute existsbound + queue
validate@tasktests / tools / human gatesvalidate + gate
C'new execution plan
Relation schema:
c → r@D,E → C'lostdegradedfallbacksubstitutedecomposevalidateboundupdateEvidence Ledger
Every recommendation should be evidence-bound. Benchmarks support capability claims; claims update task routes; task routes produce actions.
Benchmarks
- SWE-Bench Pro
- FrontierCode / long-horizon coding benchmarks
- Terminal-Bench 2.1
- OSWorld / agentic computer-use benchmarks
- Domain-specific evals
Claims
- Fable 5 unavailable changes task routing
- Large repo migration loses direct autonomous repair route
- Single-file debugging remains covered by available models
- Security-sensitive workflows require validation gates
- Fallback behavior means selected model may not always be executing model
Actions
- Use Opus 4.8 for high-precision reasoning
- Use Sonnet 4.6 for routine batch tasks
- Use GPT-5.5 where terminal/CLI integration matters
- Split large tasks into bounded chunks
- Add deterministic validators and human review gates
Task Impact Matrix
7 routes| Task | Lost / affected r | Current route | Boundary | Validation requirement |
|---|---|---|---|---|
| Single-file bug fix | locate + repair | Opus 4.8 / Sonnet 4.6 | low impact | unit testscovered |
| Multi-file refactor | locate + repair | Opus 4.8 + test harness | bounded | integration testsbounded |
| Large repo migration | repair + locate + generalize | no direct substitute | high impact | decompose + human gateno direct route |
| Security code review | validate | multi-model + human gate | high risk | manual security reviewhuman gate |
| Long PDF extraction | locate + anchor | Opus 4.8 | medium | citation verificationbounded |
| Customer support triage | select | Sonnet 4.6 | low | sampling auditcovered |
| Financial extraction | reconcile + validate | Opus 4.8 | medium/high | source reconciliationhuman gate |