dude. GPT OSS performance on a blackwell 6000 pro is INSANE.

I think the meta just changed forever.

Inconsistent gpt-oss-120b provider @chilly-france-44439 When you benchmark with GPT-OSS-120B with OpenHands, please be careful, as some providers are not reliable. link

👋 Hey all Quick question ❓ Do you want community discussions to be dropped here in #C06P5NCGSFP or would you prefer a separate channel like

#community

? Sample discussions could be reminders for community calls, community call recordings, discussion related to community engagement (good first issues, creating community demos). It's not limited to that, but just a sample. ty 1️⃣ = post in #C06P5NCGSFP 2️⃣ = create a separate

#community

channel 🤷 = don't feel strongly either way

📖 Blog post is live where @wooden-umbrella-84098 discusses the good and the bad when it comes to software agents mitigating prompt injection attacks. At the end, he shares a playbook on how to stay safe ⚾ https://www.all-hands.dev/blog/mitigating-prompt-injection-attacks-in-software-agents

https://trilogyai.substack.com/p/how-i-cut-ai-tutor-costs-by-70-with am I reading this right that genetic algorithms could be used to prompt engineer the microagent/microcontext prompts?

What's your favorite model for dealing with C#?

Hi Anyone is working on voice ai agent evaluation

Thoughts on security? https://garymarcus.substack.com/p/llms-coding-agents-security-nightmare

oh waveq Hey all Our community call is this Friday, the 22nd at 2pm (UTC-4) / 18:00 (UTC 0) We look forward to seeing you 🐕 Feel free to add our upcoming events like this one to your calendar: https://calendar.google.com/calendar/u/0?cid=Y19jNzI1MjVkZGE4ZGQ0YzZkODk4YjA0MDY1OGYyN[…]ZDgxNjhjM2VjNzQ0ZjMzMWZmQGdyb3VwLmNhbGVuZGFyLmdvb2dsZS5jb20

👋 Hello all ☑️ Time for our monthly survey 🥸 Anonymous & takes 2 minutes (or less!) https://forms.gle/Lz3PkKHdsbofwXVH7

Community Meeting is now!!! Video call link: https://meet.google.com/xvv-bdgs-snt

Recording from today's community call: • • Transcript / Notes • Chat • Presentation

Has anybody checked out the B-Mad method https://github.com/bmad-code-org/BMAD-METHOD? I'm interested to see how OpenHands can leverage it.

If you’re using github+chrome+openhands cloud and haven’t installed the openhands chrome plugin, I’d definitely recommend it: https://github.com/All-Hands-AI/openhands-chrome-extension It’s a really nice way to open new conversations on the cloud directly by pressing a button. It’s especially helpful because the repo selector on the front page of app.all-hands.dev is a little bit slow now.

Azure GPT-5 deployment error; are we using the latest litellm? https://github.com/BerriAI/litellm/issues/13432 (2 weeks ago fix appears to address this?) _BadRequestError: litellm.BadRequestError: AzureException BadRequestError - Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead._

Hi, all. I am a researcher on codegen and want to reproduce the swebench evaluation using openhands. However, I keep encountering the issue:

ERROR:root:<class 'RuntimeError'>: Maximum error retries reached for instance sympy__sympy-15809
Instance django__django-11206:   7%| | 31/465 [9:53:18<138:26:13, 1148.33s/it, Test Result: {'git_patch': 'diff --git a/django/utils/numberformat.py b/django/utils/numberformat.py\nindex 88b35fc435.

I am wondering how to proceed or who to contact for the details. I appreciate it a lot!

There are a lot of similar issues as mine, such as https://github.com/All-Hands-AI/OpenHands/issues/6287, https://github.com/All-Hands-AI/OpenHands/issues/6757, https://github.com/All-Hands-AI/OpenHands/issues/6822#issuecomment-2674002383, https://github.com/All-Hands-AI/OpenHands/issues/6287#issuecomment-2607221294, https://github.com/All-Hands-AI/OpenHands/issues/7044

Good Morning ☕ 🌅 or Evening 🌃 Our current community call is on Friday's in the Afternoon US time. I'd like to have a poll to determine if the day/time we have right now works best. Please react to the following day & times to pick the best ones for you. You're welcome to react to more than one. We will keep the cadence of every other week. ty • 🍎 Thursdays -- 11am EDT (15:00 UTC 0) • 🍊 Thursdays -- 1pm EDT (17:00 UTC 0) • 🍌 Thursdays -- 2pm (18:00 UTC 0) • 🍵 Fridays -- 11am EDT (15:00 UTC 0) • 🫐 Fridays -- 1pm EDT (17:00 UTC 0) • 🍇 Fridays -- 2pm (18:00 UTC 0) [I will ping again to resurface this later to get more eyes on it]

OpenHands got some good scores on SWT-Bench, a benchmark for generating tests for software: https://x.com/allhands_ai/status/1959964207715074421

if folks are looking for a quick way to customise the CLI system prompt I had opus write up what i am doing as a gist at https://gist.github.com/simbo1905/0b6763fa674c3c11f4ecf7a8a28c1297 #YMMV #NotLegalAdvice 😀 if anyone new likes me tries it and it glitches feel free to poke me on dm to tweak it. if any old hands know of a better way and want to provide a steer please do provide advice on the gist

🔎 New Slack Channel Nominations 👀 Since joining, I've seen requests for multiple different channels. I wanted to share some thoughts: • The #C06P5NCGSFP channel has users automatically join always. New channels may be added to default channels when new members join. But any current members will need to search to reveal & add new channels. Discovery might be difficult for current members. • So far, we've prioritized broader channels for more activity/utilization. More niche channels could be more silent than not especially due to discovery. All that being said, this Slack is to serve the community and the All Hands team. So, please discuss and request/nominate new channels in the 🧵 . Reactions & discussions to nominations will help inform us if we need a new channel or that current channels need to be broadened/clarified. So feel free to react with emojis for ones you want.

Is this a job for OpenHands Versa? https://andonlabs.com/evals/vending-bench what is the agent orchestration layer they use?

https://x.com/xai/status/1961129789944627207?t=J_DCWpCbJHRZ2UlOITJ58g&s=19 Really cheap and looks decent, are you giving it a ride on swe-bench? @chilly-france-44439

👋 Hey all! We're going to start executing on these two major initiatives next week: • https://github.com/All-Hands-AI/OpenHands/issues/10649 • https://github.com/All-Hands-AI/OpenHands/issues/10577 We're also going to be moving a large chunk of our proprietary code into the OpenHands repository. It will be in an

/enterprise

directory with its own license I don't love having two different licenses in the OpenHands repo, so we're going to try and find a way to make this better in the future. But for now it will make everyone's life much easier--especially as we migrate some of our proprietary code into the MIT-license! If you have any questions or concerns please let me know 🙂

@wooden-umbrella-84098 I would also be interested in this - https://x.com/xai/status/1961129789944627207?t=J_DCWpCbJHRZ2UlOITJ58g&s=19 Really cheap and looks decent, are you giving it a ride on swe-bench? @chilly-france-44439 X (formerly Twitter) xAI (@xai) on X Introducing Grok Code Fast 1, a speedy and economical reasoning model that excels at agentic coding. Now available for free on GitHub Copilot, Cursor, Cline, Kilo Code, Roo Code, opencode, and Windsurf. https://t.co/3tMbmLbxOP

sonnet is quite expensive and it would be good to have a cost effective option

i have tried it with current openhands but no such luck at this point

love the product btw

im hoping to share some screenshots of what im building soon 🙂