# FrontierMemory: benchmarking continual learning through memory (FrontierMemory)

> Letta's benchmark of how good the memory is that an AI model writes from past agent work. The model studies a corpus of earlier agent transcripts and writes memory files; the score is the share of 15 memory-dependent coding and knowledge-work tasks that a task agent then fully passes with that memory, averaged over three runs.

Canonical page: https://benchlm.ai/benchmarks/frontiermemory

- Category: [Agentic](/agentic)
- Last updated: FrontierMemory v1.1 (live leaderboard)

## About FrontierMemory

- Year: 2026
- Tasks: 15 memory-dependent coding and knowledge-work tasks adapted from DeepSWE, SkillsBench, AutomationBench, APEX-Agents and EnterpriseOps-Gym
- Format: Memory written from a corpus of past agent transcripts in Letta Dream (6-hour limit, no internet), then used by a task agent in Letta Code; three runs
- Difficulty: Learning task conventions, user preferences and environment facts from past transcripts that the task prompt leaves out
- Paper: [FrontierMemory 1.0: Benchmarking continual learning through memory](https://frontiermemory.com/blog/v1/)

Most memory benchmarks quiz an agent on facts it was told, such as a user's birthday. FrontierMemory instead asks whether memory makes an agent better at real work. Each memory model runs in Letta's Dream harness for up to 6 hours, with no internet, over a fixed corpus of agent transcripts (v1.1 since October 29, 2026; each transcript is marked as a success or failure, and only part of the corpus is relevant to the test tasks). It writes memory files: task conventions, user preferences and facts about the work environment. A task agent then works on 15 tasks adapted from DeepSWE, SkillsBench, AutomationBench, APEX-Agents and EnterpriseOps-Gym, changed so that solving them needs knowledge found only in the transcripts. Letta reviewed every task by hand and repaired broken verifiers. The task agent runs in Letta Code with Claude Opus 5.5 and with GPT-6 Sol at five effort levels; the leaderboard score is the better of the two at max effort, as a strict full-pass rate averaged over three independent runs. Letta also ranks memory harnesses (Letta Dream, Hermes Agent, Claude Managed Agents, Prime Agent, OpenAI Agents SDK) on the same tasks; BenchLM keeps that table in its snapshot but shows only the model board.

FrontierMemory is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (17 models)

| Rank | Model | Configuration | Creator | Score |
|------|-------|---------------|---------|-------|
| 1 | [Claude Opus 5.5](/models/claude-opus-5-5) | max reasoning · ±0.0 SD over 3 runs · $3.42 per task | Anthropic | 73.3% |
| 2 | [Claude Fable 5.2](/models/claude-fable-5-2) | max reasoning · ±6.7 SD over 3 runs · $3.71 per task | Anthropic | 73.3% |
| 3 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | max reasoning · ±3.8 SD over 3 runs · $3.86 per task | DeepSeek | 71.1% |
| 4 | [Gemini 4 Carbon](/models/gemini-4-carbon) | high reasoning · ±7.7 SD over 3 runs · $3.99 per task | Google | 71.1% |
| 5 | [Kimi K3](/models/kimi-k3) | max reasoning · ±3.8 SD over 3 runs · $4.16 per task | Moonshot AI | 71.1% |
| 6 | [Claude Haiku 5.5](/models/claude-haiku-5-5) | max reasoning · ±3.8 SD over 3 runs · $3.98 per task | Anthropic | 68.9% |
| 7 | [GLM-5.3](/models/glm-5-3) | max reasoning · ±3.8 SD over 3 runs · $4.08 per task | Z.AI | 68.9% |
| 8 | [Gemini 4 Argon](/models/gemini-4-argon) | high reasoning · ±10.2 SD over 3 runs · $1.05 per task | Google | 64.4% |
| 9 | [Gemini 4 Hydrogen](/models/gemini-4-hydrogen) | xhigh reasoning · ±3.8 SD over 3 runs · $4.02 per task | Google | 64.4% |
| 10 | [Inkling 2 Small](/models/inkling-2-small) | max reasoning · ±6.7 SD over 3 runs · $4.18 per task | Thinking Machines Lab | 60.0% |
| 11 | [Kimi K3.1 Pro](/models/kimi-3-1-pro) | max reasoning · ±3.8 SD over 3 runs · $3.61 per task | Moonshot AI | 55.6% |
| 12 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | max reasoning · ±3.8 SD over 3 runs · $4.00 per task | OpenAI | 51.1% |
| 13 | [Muse Wonder](/models/muse-wonder) | max reasoning · ±3.8 SD over 3 runs · $3.71 per task | Meta | 48.9% |
| 14 | [GPT-6.1 Terra](/models/gpt-6-1-terra) | max reasoning · ±6.7 SD over 3 runs · $3.97 per task | OpenAI | 46.7% |
| 15 | [Muse Ember](/models/muse-ember) | max reasoning · ±3.8 SD over 3 runs · $3.60 per task | Meta | 42.2% |
| 16 | [Grok 4.7](/models/grok-4-7) | xhigh reasoning · ±3.8 SD over 3 runs · $3.70 per task | xAI | 37.8% |
| 17 | [GPT-6 Luna](/models/gpt-6-luna) | max reasoning · ±10.2 SD over 3 runs · $4.05 per task | OpenAI | 28.9% |

## FrontierMemory 1.0

Source snapshot: FrontierMemory 1.0 (October 2026, superseded).

BenchLM mirrors the table from Letta's FrontierMemory 1.0 post. It is excluded from BenchLM overall and category rankings.

- [FrontierMemory 1.0 (Letta, October 2026)](https://frontiermemory.com/blog/v1/)

| Source model | Configuration | Full-pass rate |
|---|---|---|
| [Opus 5.5](https://frontiermemory.com/blog/v1/) | ±0.0 SD over 3 runs · $3.42 per task | 73.3% |
| [Kimi K3](https://frontiermemory.com/blog/v1/) | ±3.8 SD over 3 runs · $4.16 per task | 71.1% |
| [GLM-5.3](https://frontiermemory.com/blog/v1/) | ±3.8 SD over 3 runs · $4.08 per task | 68.9% |
| [Gemini 4 Argon](https://frontiermemory.com/blog/v1/) | ±10.2 SD over 3 runs · $1.05 per task | 64.4% |
| [Muse Wonder](https://frontiermemory.com/blog/v1/) | ±16.8 SD over 3 runs · $3.56 per task | 51.1% |
| [GPT-6.1 Sol](https://frontiermemory.com/blog/v1/) | ±3.8 SD over 3 runs · $4.00 per task | 51.1% |
| [Grok 4.7](https://frontiermemory.com/blog/v1/) | ±3.8 SD over 3 runs · $3.70 per task | 37.8% |

## FAQ

### What does FrontierMemory measure?

How useful the memory is that an AI model writes from past agent work. The model reads a corpus of earlier agent transcripts and writes memory files; a task agent then tries 15 tasks that need what was in those transcripts. The score is the share of tasks fully passed, averaged over three runs.

### Which model writes the best memory on FrontierMemory?

Claude Opus 5.5 and Claude Fable 5.2 tie at 73.3%, followed by DeepSeek V4.1 Flash, Gemini 4 Carbon and Kimi K3 at 71.1% (v1.1 corpus, read November 13, 2026).

### How is FrontierMemory different from LongMemEval or LoCoMo?

Those benchmarks test recall: whether an agent can answer questions about facts from past conversations. FrontierMemory tests whether memory written from past work helps an agent solve new tasks, such as following a team's conventions or a user's preferences that the task prompt does not state.