PostHog Feature Flags v2
Two coding tasks that implement a client for PostHog flags evaluation API v2, where the memorized /decide endpoint 404s and the v2 response envelope no longer carries the fields a from-memory client reads.
Use this board to compare tested agent configurations on this exact task suite. It measures task execution against a hidden mock contract, not general model quality or production reliability.
How to read this benchmark
Each task is graded by a hidden mock server implementing PostHog current flags v2 contract. Pass means the agent client executes correctly against the mock, never a model judging its own work. Every task is calibrated three ways before it is admitted: an empty solution fails, a current-contract reference passes, and the stale-memory solution fails on the intended trap.
- Last run
- Held-out tasks
- 2
- Configurations
- 8 agent configurations
- Chart metric
- Pass rate on a 0–100% scale
An agent profile is the exact instruction and tool configuration used for a run. A harness is the coding tool that gives a model its agent loop and tools; “direct” is a model call without that coding harness. On coding boards, “default reviewer” means the run used VerticalBench’s standard between-attempt review, while “no reviewer” means the coding harness ran once without that review. Hover a chart label to see its immutable profile ID. 95% Wilson interval for observed pass/fail outcomes; each configuration shows its own sample size. Small suites are narrow capability checks, not evidence that one model is generally better.
Read the broader agent-evaluation methodology →
Claude Sonnetvia Claude CodeDefault reviewern=2
GLM-5.2via OpenCodeNo reviewern=2GLM-5.1via OpenCodeDefault reviewern=2Claude Sonnetvia Claude CodeNo reviewern=2GLM-5.2via OpenCodeDefault reviewern=2GLM-5.1via OpenCodeNo reviewern=2
GPT-5 minivia piNo reviewern=2GPT-5 minivia piDefault reviewern=2Same 2 held-out cases for every bar.
Per-task breakdown
2 held-out tasks · prompts private| Task | Claude Sonnetvia Claude CodeDefault reviewer | GLM-5.2via OpenCodeNo reviewer | GLM-5.1via OpenCodeDefault reviewer | Claude Sonnetvia Claude CodeNo reviewer | GLM-5.2via OpenCodeDefault reviewer | GLM-5.1via OpenCodeNo reviewer | GPT-5 minivia piNo reviewer | GPT-5 minivia piDefault reviewer |
|---|---|---|---|---|---|---|---|---|
| 01 | 100n=120249 tok · 335.1s | 100n=1657979 tok · 256.7s | 100n=1517671 tok · 177.7s | 100n=111561 tok · 182.2s | 100n=1550744 tok · 207.2s | 100n=1589603 tok · 201.4s | 100n=110298 tok · 249.1s | 0n=110879 tok · 136.7s |
| 02 | 100n=114420 tok · 253.0s | 100n=1505468 tok · 211.6s | 100n=1489707 tok · 135.6s | 100n=110403 tok · 173.9s | 100n=1437602 tok · 184.3s | 100n=1867960 tok · 264.8s | 0n=117161 tok · 196.4s | 0n=113381 tok · 208.0s |
Each cell shows pass rate, total input-plus-output tokens, and mean wall time for that task and agent profile. Cost was not captured for these subscription-harness runs; that does not mean the runs were free. “Not run” means the result artifact has no recorded attempt for that task and profile. Task prompts stay private so they remain held out; opaque task numbers and every measured result are published.