[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.] [@silasalberti](/creator/twitter/silasalberti) "super excited to release SWE-grep and SWE-grep-mini SWE-grep-mini achieves extreme inference speeds of 2800 TPS: 20x faster than Haiku XXX while beating Sonnet XXX Opus XXX & GPT-5 on our CodeSearch eval our vision: make agentic search as fast as embedding search. the SWE-grep models can perform multi-turn agentic search in X seconds compared to 20-60 seconds for other frontier models" [X Link](https://x.com/silasalberti/status/1978871477605929229) [@silasalberti](/creator/x/silasalberti) 2025-10-16T17:11Z 11.5K followers, 64.6K engagements "one key thing that we optimized for is making the model a good companion for any smart frontier model Sonnet XXX in Windsurf can call SWE-grep as a tool. on Codebase Q&A tasks we achieve 2.7x speed-ups compared to Claude Code 4.8x compared to Cursor CLI" [X Link](https://x.com/silasalberti/status/1978871480093151705) [@silasalberti](/creator/x/silasalberti) 2025-10-16T17:11Z 11.5K followers, 1415 engagements "it turns out that with very fast inference 50+% of the end-to-end latency isnt the model anymore but network & tool call overhead. each additional serial agent turn is costly we find that X serial turns + X parallel tool calls outperforms X serial turns + X parallel tool calls. during our RL training we saw nice emergent rise in parallel tool calling. more details on how we achieved this in our blog post" [X Link](https://x.com/silasalberti/status/1978871483738001850) [@silasalberti](/creator/x/silasalberti) 2025-10-16T17:11Z 11.5K followers, 1055 engagements "@CerebrasSystems @windsurf ⚡" [X Link](https://x.com/silasalberti/status/1978916338279280946) [@silasalberti](/creator/x/silasalberti) 2025-10-16T20:09Z 11.5K followers, XXX engagements "check out our blog post with details about training here: try it side-by-side against other agents in our playground: or you can try it in Windsurf Agent:" [X Link](https://x.com/silasalberti/status/1978871486531383514) [@silasalberti](/creator/x/silasalberti) 2025-10-16T17:11Z 11.5K followers, 1184 engagements "Try out our Fast Context playground that allows you to compare SWE-grep against Claude Code: It's back up after temporary capacity issues yesterday night" [X Link](https://x.com/silasalberti/status/1979329799282925674) [@silasalberti](/creator/x/silasalberti) 2025-10-17T23:32Z 11.5K followers, 10.8K engagements
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
@silasalberti
"super excited to release SWE-grep and SWE-grep-mini SWE-grep-mini achieves extreme inference speeds of 2800 TPS: 20x faster than Haiku XXX while beating Sonnet XXX Opus XXX & GPT-5 on our CodeSearch eval our vision: make agentic search as fast as embedding search. the SWE-grep models can perform multi-turn agentic search in X seconds compared to 20-60 seconds for other frontier models"
X Link @silasalberti 2025-10-16T17:11Z 11.5K followers, 64.6K engagements
"one key thing that we optimized for is making the model a good companion for any smart frontier model Sonnet XXX in Windsurf can call SWE-grep as a tool. on Codebase Q&A tasks we achieve 2.7x speed-ups compared to Claude Code 4.8x compared to Cursor CLI"
X Link @silasalberti 2025-10-16T17:11Z 11.5K followers, 1415 engagements
"it turns out that with very fast inference 50+% of the end-to-end latency isnt the model anymore but network & tool call overhead. each additional serial agent turn is costly we find that X serial turns + X parallel tool calls outperforms X serial turns + X parallel tool calls. during our RL training we saw nice emergent rise in parallel tool calling. more details on how we achieved this in our blog post"
X Link @silasalberti 2025-10-16T17:11Z 11.5K followers, 1055 engagements
"@CerebrasSystems @windsurf ⚡"
X Link @silasalberti 2025-10-16T20:09Z 11.5K followers, XXX engagements
"check out our blog post with details about training here: try it side-by-side against other agents in our playground: or you can try it in Windsurf Agent:"
X Link @silasalberti 2025-10-16T17:11Z 11.5K followers, 1184 engagements
"Try out our Fast Context playground that allows you to compare SWE-grep against Claude Code: It's back up after temporary capacity issues yesterday night"
X Link @silasalberti 2025-10-17T23:32Z 11.5K followers, 10.8K engagements
/creator/twitter::485974393/posts