SuperAgent API
Memory benchmarks
What SuperAgent API memory is measured to do, on public benchmarks, with every answer published.
Two surfaces share one memory design: hosted memory (the Cloud Superbrain behind this API’s memory.* operations) and the Hive Superbrain that runs on your own computer (hive-brain answer). They recall the same way; the hosted Memory Answer works within one 24,000-character budget, while the computer version can hand over up to 60,000. Each result below says which one it measures.
Hosted memory (this API)
| Measure | Result |
|---|---|
| Evidence reaches the reading model, LongMemEval_S (each answer among ~50 past chats) | 90.6% |
| Evidence reaches the reading model, LongMemEval Oracle (only the chats that hold the answer) | 98.5% |
| Answers correct, 150 LongMemEval_S questions | 81.9% |
Evidence: all 500 questions (470 answerable), memory.answer in keyword order, no model calls. Answers: the shipped reader, graded by LongMemEval’s own judge prompts with gpt-4o-mini, measured on 2026-09-23 and not yet re-run on the current evidence layout.
Hive Superbrain (same design, on your computer)
Claude Opus 4.6 reads what memory returns; LongMemEval’s official judge prompts with gpt-4o-2024-08-06 grade every answer.
| Benchmark | Hive Superbrain | Compared with |
|---|---|---|
| LongMemEval Oracle, 470 questions | 96.6% | Sibyl-Memory’s published Opus 4.6 answers, graded the same way: 95.3% (p = 0.33) |
| LongMemEval_S, 470 questions | 94.3% | Sibyl-Memory’s published architecture, run with its public runner: 92.1% (p = 0.16), reading about a ninth of the tokens |
| LoCoMo, 500 questions | 94.4% | A pre-registered held-out test: none of the memory design was tuned on LoCoMo |
HivemindOS scored higher than Sibyl-Memory on both comparisons; neither difference is established beyond chance.
Receipts
Every context memory returned, every answer and every verdict is published in hivemindos-memory-benchmarks, with scripts that recompute each number from the files. The full method, including what was tried and did not help, is on the benchmarks page. How to call hosted memory is in Memory.