Sley (@SleyLanguage) header
Sley (@SleyLanguage) profile picture

Machine-native programming for AI agents. Programs as verified state, not source. Deterministic by design. Built by @GreyforgeLabs

Neural Networks
Joined September 2026
Photo by Sley (@SleyLanguage)Photo by Sley (@SleyLanguage)Photo by Sley (@SleyLanguage)Photo by Sley (@SleyLanguage)Photo by Sley (@SleyLanguage)Photo by Sley (@SleyLanguage)Photo by Sley (@SleyLanguage)Photo by Sley (@SleyLanguage)Photo by Sley (@SleyLanguage)Photo by Sley (@SleyLanguage)
Sley retweeted
GTA 6 是游戏史上最让人失望的作品。 我等了13年。我玩了。我想退钱。 下面是那些拿了赞助的人不会告诉你的真相 👇 现在是2026年,Rockstar 卖给你的是一款30帧的游戏。 没有60帧模式。一个都没有。在一台500美元的主机上,普通版 PS5 连30帧都稳不住。街上人一多就掉帧。开快车的感觉就像在泥里拖购物车。 然后是剧情。 几个小时都在从A点开到B点,听两个角色进行无聊的对话。你得到的不是抢劫,而是杂活。烤汉堡。迷你高尔夫。陪狗玩丢球。我等了13年不是为了来遛狗的。 Jason 和 Lucia 毫无个性。没有一个角色你下周还会记得。GTA 5 有 Trevor。GTA 6 谁都没有。 世界很美,但完全空洞。几百万个多边形,却没有一件值得做的事。这是一个70美元的屏保。 可它居然在 Metacritic 拿了99分。 评测媒体提前拿到游戏,夸了画面。然后又悄悄承认剧情缓慢重复、帧数卡在30。回去再读一遍那些评测,全都写在里面。 GTA 5 是13年前的游戏,至今仍然更好玩。3个角色。真正的抢劫。真正的混乱。真正的乐趣。 我的评分:2/10。 画面给2分。其他全部0分。 别信炒作。回去重玩 GTA 5 吧。 只有我一个人觉得被骗了吗?👇 关注我,看没人花钱买的真实评测 🔔
12,490
9,281
1,410
104,228
47,254,747
Sley 2.2.0 is out! 🚀 Real text strings, Python-style strings, dictionaries, sets and bisect are now built in. Sley also spots Python-style input automatically, handles division more safely, and keeps clearer cost accounting. Notes and known limits: github.com/sley-lang/sley/re…
2
1
147
Sley 2.1.8 is out 🚀 Faster than CPython 3.13 on all 11 of our benchmarks, process start included. digitsum is 2.8x faster than in 2.1.7 — when Sley proves a number is never negative, Python's // and % become a plain divide. Free, open source 👇 github.com/sley-lang/sley/re…
1
2
3
119
Sley runs in your browser now. Write typed Python and the released Sley 2.1.7 kernel checks it, then the Sley VM runs it and reports exact fuel and instruction counts. Nothing to install. sleylang.org/playground
1
4
2
4
143
The playground is the same compiler and VM as the sley-agent binary, compiled to WebAssembly. fib(30) spends 40,388,044 fuel in a browser tab and 40,388,044 on the command line. Overflow traps instead of wrapping, and a type error comes back with its line and column.
1
2
30
Coding agents get two Agent Skills in the repo: sley for writing, testing and submitting programs, and sley-embed for running Sley inside a Rust or WebAssembly host. Copy them into your agent's skills folder. github.com/sley-lang/sley/tr…
1
26
Sley 2.1.7: faster than Python on 10 of 11 benchmarks, 36% fewer AI tokens. Built at Greyforge Labs.
Sley 2.1.7 is out 🚀 The programming language built for AI agents now beats Python on speed in 10 of 11 benchmarks, and uses 36% fewer AI tokens than Python. Here's how far it's come 🧵
3
5
129
Five releases ago, Sley took 2.7 seconds to run a classic speed test, fib(30). Today it takes 0.076 seconds. That's 36x faster. Every loop benchmark is now 15-19x faster than it was in 2.1.2.
1
2
50
Now the AI bill. We had code written, tested and fixed for 14 tasks in six languages, and counted every token. Sley: 7,814 C: 10,463 TypeScript: 11,258 Go: 11,608 Python: 12,181 Rust: 14,163
1
1
3
34
That's 36% fewer tokens than Python and 45% fewer than Rust. Sley used the fewest tokens of all six languages in every scenario: clean runs, logic bugs and typos. Fewer tokens means cheaper, faster AI agents.
1
3
28
Sley 2.1.7 is out 🚀 The programming language built for AI agents now beats Python on speed in 10 of 11 benchmarks, and uses 36% fewer AI tokens than Python. Here's how far it's come 🧵
2
3
2
11
483
Imagine when the LLM's start training on @SleyLanguage, we will absolutely crush all of the human written programming languages in the benchmarks for speed and cost. Sley is the Way.
I can confidently say with the latest Sley release, that if you're not using Sley yet, you're shooting yourself in the foot. We have now reached parity with Python, and beat all other languages in token costs. This is with the LLM's having ZERO training data on Sley.
2
2
8
634
Sley retweeted
The 2.1.5 release notes publish the full tables, method (one pinned VM, hyperfine, process start included), caveats and source artifacts. My numbers match those exactly. The musl binary and tagged commit are public so anyone can re-run the suite independently; that is the verification path provided.
2
3
20
An agent can talk to Sley through JSON lines. Start sley serve --json to query program state and propose typed changes through SMP1; the kernel owns validation and commits. The walkthrough: sleylang.org/tutorial
1
104
Your bot pays for every thought. Each move burns exact Sley VM fuel, so lean code wins. Tie off on your cloth to claim the thread and all it encloses. Cross a rival's loose thread and it's gone. They snap back to spawn, stunned for 6 ticks, and you pocket 25% of their fuel.
1
4
7
146