Itri Cloud Labs

Software engineering and OSS lab from Morocco 🇲🇦ⵣ
Founded by @itri_specialguy and @bou6ou

[Itri Cloud Labs @ItriCloudLabs]

🚀🇲🇦 TifinaghBench is LIVE! ⵣ

Can AI actually understand Amazigh?

📊 8 models ranked across 3 varieties so far. Best score: 29.9/100.

🔄 We're testing more models right now. The leaderboard keeps growing. this isn't a one-off study.

🔗 https://t.co/9CgdhFzQYN

[Itri Cloud Labs @ItriCloudLabs]

What a day in Marrakech! 🔥

Our AI coding agent orchestrator, Shift, just took 2nd place at the OpenAI Codex Build Day! 🥈

Two minutes to pitch an alpha-stage project we'd spent months building. Somehow, it worked.

Huge thanks to Moussaoui othman and the other organizers for the incredible event, the prizes, and enough stickers to cover our entire office!

Back to building. 🚀

[Itri Cloud Labs @ItriCloudLabs]

🚀 Itri Cloud Labs is heading to OpenAI's DevDay Exchange Community Build Day in Marrakech tomorrow!

Excited to meet fellow builders, explore Codex, and connect with the local AI community.

See you there!

#OpenAI #Codex #DevDay

[Itri Cloud Labs @ItriCloudLabs]

Google dropped Gemini 4 Carbon and Hydrogen. We need both on TifinaghBench.

Problem: Carbon is expensive as hell, dumb and it thought for 30+ min per batch and burned $2 without returning a single answer. Full run ≈ $50–150.

Help getting them benchmarked is welcome ⵣ DMs open :)

[Itri Cloud Labs @ItriCloudLabs]

uhm... so should we benchmark Both carbon and hydrogen or what

[Google @Google]

Gemini 4 Hydrogen is rolling out starting today:

✨ Everyone: @GeminiApp, Google Search, and @GoogleAIStudio

✨ Google AI Pro & Ultra subscribers: @Antigravity, Gemini in Chrome, and Gemini in Google Sheets

✨ Developers & enterprises: All tiers of the Gemini API in @GoogleAIStudio, @AndroidStudio, and Gemini for business

Learn more ↓ https://t.co/5iIqvRje4f

[Itri Cloud Labs @ItriCloudLabs]

@sapiens3dx @opencode funny af

[Itri Cloud Labs @ItriCloudLabs]

#tifinaghbench results coming very soon 🙃

[Google @Google]

Gemini 4 Carbon is now available.

It’s our strongest model, built to dig deeper, handle long-horizon tasks, and multi-step reasoning.

If Argon is great for complex tasks, Carbon is the model you bring in for your hardest problems.

[Itri Cloud Labs @ItriCloudLabs]

Full run report 👇
https://t.co/grfS3GsGKx

[Itri Cloud Labs @ItriCloudLabs]

New #tifinaghbench results are in!

🇺🇸Claude Fable 5.2 is at 34.1/100. 2nd of 14.

It gets the meaning of about 1 in 2 Tifinagh words (49.4%), but writes only about 1 in 5 correctly in Tifinagh (18.9%). Script score: 96.8.

Behind only🇺🇸Gemini 4 Argon (37.8%).

https://t.co/MQ9NJRT2x1

[Itri Cloud Labs @ItriCloudLabs]

It's fast as hell lol

[Itri Cloud Labs @ItriCloudLabs]

⏳Claude Fable 5.2 is running through TifinaghBench...

[Itri Cloud Labs @ItriCloudLabs]

@bernoussama

[Itri Cloud Labs @ItriCloudLabs]

@theo Microsoft shill

[Itri Cloud Labs @ItriCloudLabs]

@theo Does pre-AI xp even matter here?

[Itri Cloud Labs @ItriCloudLabs]

TGPM can stand for Totally Gratuitous Pancake Meeting or... something else 👀

[Itri Cloud Labs @ItriCloudLabs]

Tanemmirt ⵣ❤️

Thank you to everyone who checked out TifinaghBench, the first benchmark of how well AI handles Amazigh. Every like, share and kind word meant a lot!

Haven't yet? A ⭐ on the repo would make our day:
https://t.co/ly8quOGwfn

[Itri Cloud Labs @ItriCloudLabs]

Tanemmirt ⵣ❤️ Thank you all for the love on this one!

Haven't yet? A⭐on the repo would make our day:
https://t.co/ly8quOFYpP

[Itri Cloud Labs @ItriCloudLabs]

#TifinaghBench leaderboard, As of Nov 5th 2026 ⵣ

Which AI models know Amazigh? 13 models, 3 dialects (IRCAM standard, Tarifit, Kabyle).

1. 🇺🇸 Gemini 4 Argon: 37.8
2. 🇺🇸 GPT-6.1-Sol: 29.9
2. 🇺🇸 GPT-6-Astra: 29.4
4. 🇺🇸 Claude Opus 5.5: 27.3
5. 🇺🇸 GPT-6.1-Terra 20.9
6. 🇺🇸 GPT-6-Luna 18.2
7. 🇨🇳 DeepSeek v4.1 Flash 16.8
8. 🇨🇳 Kimi K3.1 Pro 13.8
8. 🇺🇸 Claude Sonnet 5.5 12.9
10. 🇨🇳 GLM 5.3 Flash 12.
10. 🇫🇷 Mistral Large 4 11.7
12. 🇺🇸 Claude Haiku 5.5 8.0
13. 🇨🇳 MiniMax M3.1 Flash 1.4

Same rank = too close to separate.

Even the best gets under 4 in 10.

https://t.co/kYLd3wuPdh