Model economics / detail
gpt-5.6-solC
Dispatch tier: Expensive (complexity class this model is routed for)
136 runs is a real sample, and 0.43 is a real problem — worst rate of any high-volume model. The merge-conflict beat is genuinely brutal, which buys it a C instead of a D, but hazard pay isn't a grade.
The measured record
What the ledger shows
Fleet-ledger aggregates for this model only — run counts, outcomes, spend, and token appetite. Missing telemetry reads “Not observed”, never zero.
Agent runs
136
Success rate
43%
Metered spend
Not observed
cost-blind: no run was metered
Cost / metered run
Not observed
Tokens in
Not observed
Tokens out
Not observed
Tokens in / run
Not observed
Tokens out / run
Not observed
Outcomes
Status breakdown
Every run ends in exactly one status. The bar is the whole record, to scale.
- failure: 72 of 136 runs (52.9% of total)
- killed: 3 of 136 runs (2.2% of total)
- timeout: 1 of 136 runs (0.7% of total)
- lost: 2 of 136 runs (1.5% of total)
Run duration
How long the runs take
Wall-clock distribution across measured runs, from the fastest exit to the longest grind.
Min
6s
p25
13s
Median
5m
p75
22m
p90
35m
p99
2.2h
Max
3.1h
Mean
15m
Duration measured on 136 of 136 runs.
Commentary
Idiosyncrasies
What stands out in this model's numbers — shape, appetite, and failure habits.
- The only high-volume model that fails more than it succeeds: 72 failures to 58 successes across 136 runs (43%).
- Widest duration spread in the fleet: p25 of 13 seconds, p99 of 2.2 hours, max over 3 hours.
- Completely cost- and token-blind — 136 runs and not one metered dollar or counted token in the record.
Commentary
Lessons learned
Practical routing and operations takeaways, grounded in the same record.
- 0.43 at n=136 is signal, not variance — stop routing it expensive-tier work by default.
- Its true cost is unknowable from this record; a model this cost-blind and failure-prone is double jeopardy for the budget.
- The 13-second p25 says a chunk of runs die on arrival — worth checking the harness before the next batch.