Benchmarks · Agents & computer use · Long documents
FrontierMemory v1.1
How useful the memory a model writes is. It reads trajectories of past agent work and writes memory; a task agent (mostly Opus 5.5 at max, in Letta Code) then solves new tasks using it. The share of tasks solved, mean of 3 runs. From Letta.
17 models · 17 results · data as of 2026-11-13 · frontiermemory.com ↗
Leaderboard
Fable 5.2 max
More from this benchmark
2 published measuresScore vs any published measure2 measures: sub-scores, splits, counts
5 more models under All models in settings.
All results
17 rows| # | Model | Effort | FrontierMemory v1.1 | Harness |
|---|---|---|---|---|
| 1 | Fable 5.2 | max | 73.3% | Letta Dream |
| 2 | Opus 5.5 | max | 73.3% | Letta Dream |
| 3 | DeepSeek V4.1 Flash | max | 71.1% | Letta Dream |
| 4 | Gemini 4 Carbon | high | 71.1% | Letta Dream |
| 5 | Kimi K3 | max | 71.1% | Letta Dream |
| 6 | Haiku 5.5 | max | 68.9% | Letta Dream |
| 7 | GLM 5.3 | max | 68.9% | Letta Dream |
| 8 | Gemini 4 Hydrogen | xhigh | 64.4% | Letta Dream |
| 9 | Gemini 4 Argon | high | 64.4% | Letta Dream |
| 10 | Inkling 2 Small | max | 60.0% | Letta Dream |
| 11 | Kimi K3.1 | max | 55.6% | Letta Dream |
| 12 | GPT-6.1 Sol | max | 51.1% | Letta Dream |
| 13 | Muse Wonder | max | 48.9% | Letta Dream |
| 14 | GPT-6.1 Terra | max | 46.7% | Letta Dream |
| 15 | Muse Ember | max | 42.2% | Letta Dream |
| 16 | Grok 4.7 | xhigh | 37.8% | Letta Dream |
| 17 | GPT-6 Luna | max | 28.9% | Letta Dream |