Letta (@Letta_AI) header
Letta (@Letta_AI) profile picture

Letta

@Letta_AI

Building machines that learn.

 npm i -g @letta-ai/letta-code
letta.com
Joined August 2024
  • Tweets848
  • Following0
  • Followers10.8K
Introducing FrontierMemory: a new way to evaluate agent memory through downstream task performance. While existing memory benchmarks focus on recall (saturated via agentic search), FrontierMemory focuses on how agents learn from experience.
Image from Letta (@Letta_AI)
43
35
27
283
61,428
Small models can dream too. We just added Meta’s new Muse Ember model to the FrontierMemory leaderboard, which measures an agent’s ability to self-improve through dreaming. Muse Ember performs similarly to Muse Wonder at a fraction of the cost.
Image from Letta (@Letta_AI)
2
1
1
16
1,555
Among the small models, Haiku 5.5 performs the best, nearing Opus 5.5 quality but at under a third of the cost per dream. GPT-6 Luna struggles to generate good memories, trailing at the bottom. Ember sits in-between the two, while being significantly cheaper than Luna.
Image from Letta (@Letta_AI)
1
2
207
Claude Fable 5.2 ties Opus for #1 on FrontierMemory. The quality of Fable 5.2's generated memories beat Opus 5.5 at every lower effort level. It's pricier to dream with, costing ~$600 per dream compared to ~$300 for Opus, 2x the cost.
Image from Letta (@Letta_AI)
6
2
18
4,342
See the full leaderboard at frontiermemory.com
FrontierMemory

FrontierMemory

Benchmarking continual learning through memory.

frontiermemory.com
1
258
Inkling 2 Small from @thinkymachines scores 60.0% on FrontierMemory, the highest of any US open-weights model, at about $19 per (American) dream.
Image from Letta (@Letta_AI)
3
1
1
24
1,803
Congrats to the @thinkymachines team on the model release, and upcoming Apache 2.0 weights release! For the full leaderboard, visit frontiermemory.com
FrontierMemory

FrontierMemory

Benchmarking continual learning through memory.

frontiermemory.com
3
401
Understanding how different models learn from agent experience

Understanding how different models learn from agent experience

Article · @Letta_AI
5
11
3
93
7,417
We've updated FrontierMemory to v1.1, with a larger 750-trajectory corpus that reduces reward hacking and includes a more balanced mix of sources. Opus 5.5 remains the strongest memory model, with open-weight models like DeepSeek V4.1 Flash, Kimi K3, and GLM-5.3 close behind.
Image from Letta (@Letta_AI)
8
4
19
1,878
Across harnesses, Letta Dream leads the Pareto frontier, with Hermes Agent second at higher efforts and Prime Agent second at lower ones.
Image from Letta (@Letta_AI)
1
3
528
FrontierMemory is now on @FireworksAI_HQ's Specialized Intelligence Index. Picking a model for a long-running agent? You can now compare how well each one learns from its own past attempts, combined with cost. View FrontierMemory results under RSI: fireworks.ai/specialized-int…
Image from Letta (@Letta_AI)
This quoted post is unavailable.
Try viewing anyway · View on X
2
2
1
24
1,789
We've added GPT-6.1 Terra, Kimi K3.1 Pro and DeepSeek V4.1 Flash to the FrontierMemory leaderboard. Kimi K3.1 Pro ties Opus 5.5 and GLM-5.3 for #1, and DeepSeek V4.1 Flash beats GPT-6.1 Sol and Gemini 4 Argon, for a fraction of the cost.
16
1
1
34
3,450
Full leaderboard with cost analysis at: frontiermemory.com
FrontierMemory

FrontierMemory

Benchmarking continual learning through memory.

frontiermemory.com
2
357
> GPT-6.1 Terra is now available in @Letta_AI
2
2
1
17
1,351
You can now run FrontierMemory to measure how well your model or harness improves with experience. We’re also now accepting submissions to the official leaderboard. Code and submission instructions are available at: github.com/letta-ai/frontier…
GitHub - letta-ai/frontiermemory: FrontierMemory: benchmarking continual learning through memory

GitHub - letta-ai/frontiermemory: FrontierMemory: benchmarking continual learning through memory

FrontierMemory: benchmarking continual learning through memory - letta-ai/frontiermemory

github.com
4
13
1
64
3,618
Honored to have been a featured ecosystem partner by @JensenHuang at GTC Berlin! We're excited to partner with @NVIDIA on research towards developing agents that continually learn and self-improve.
Image from Letta (@Letta_AI)
4
5
1
33
2,658
Letta Code now supports dynamic workflows to compose model-agnostic subagents and decision models.
Dynamic workflows for stateful agents and decision models

Dynamic workflows for stateful agents and decision models

Article · @Letta_AI
9
1
4
29
11,967
Opus 5.5, GPT 6 Sol, and GPT 6 Luna are now all available on Letta Code!
2
2
9
1,707
We’re releasing the Letta Agent SDK: A TypeScript library for working with stateful agents. Persistent agents can be invoked across computers (local/remote) while retaining memory and state. Use it to create your own stateful agent UIs or to build dynamic workflows.
Image from Letta (@Letta_AI)
21
3
2
54
7,816
The Letta Agents SDK is built on top of the same WebSockets interface as Letta Desktop. You can create your own custom interfaces (a web app, a mobile app, a kanban board) that connect to either Letta Cloud or your own self-deployed server.
Image from Letta (@Letta_AI)
1
3
1,455
Grok 4.6 and DeepSeek V4 Flash are now available on Letta Cloud. Use your Letta API credits to run either model.
23
2,263
How well do frontier models use memory?

How well do frontier models use memory?

Article · @Letta_AI
2
4
2
61
10,713
Access stateful agents through any OpenAI-compatible interface. Run `letta server --openai-api` to chat with stateful agents through @OpenWebUI @LibreChatAI @jandotai & more.
1
1
1
8
2,289
Opus 5 is now available in Letta! In our internal evaluations, we found that Opus 5: - Matches Fable in memory generation (rewriting memory and learning skills) - Significantly improves over Opus 4.8 in memory adherence and memory/skill usage
Claude (@claudeai) profile pictureClaude
@claudeaiJul 24
Introducing Claude Opus 5. It's a thoughtful and proactive model that comes close to the frontier intelligence of Fable 5 at half the price.
1
13
3,084
Trajectory: A Standard Format for Agent Experience Data

Trajectory: A Standard Format for Agent Experience Data

Article · @Letta_AI
31
55
25
745
209,824
Office Hours are today! Come join the livestream at 11:30am Pacific. New stuff: - Cloud sandboxes are now persistent -- one per agent - GitHub integration -- clone repositories into an agent's cloud sandbox - Schedule management on Letta Chat - GPT 5.6/Grok 4.5 support Link below:
1
3
1,133
Join here: discord.gg/JpPhc5BZB?event=1…
Discord - Group Chat That’s All Fun & Games

Discord - Group Chat That’s All Fun & Games

Discord is great for playing games and chilling with friends, or even building a worldwide community. Customize your own space to talk, play, and hang out.

discord.com
1
559
Load more