🚀 Introducing SaulLM-141B and SaulLM-54B: The First Open Family of Legal Models. After #SaulLM-7B the family is growing! We are proud to unveil the latest innovations from our team: the SaulLM-141B and 54B generative AI models, specifically designed for the legal domain.

Awesome to see them build on the same tricks as BidirLM-Omni, like the decoder-to-encoder adaptation and cross-modality model merging. The synthetic data ablation is also a huge highlight with a 15 gain for code retrieval domains

Gemini Embedding 2 is out, and it's completely omnimodal 😎 Great to see the next chapter of encoders heading toward text, audio, and visual unlocking so many use cases arxiv.org/abs/2605.27295

👏 Congratulations to @cohere on Command A — a powerful new model optimized for NVIDIA Blackwell and trained using NVIDIA CUDA-X libraries. Proud to be a part of it! Learn more ⤵️

Releasing open-source under the Apache 2.0 license. We want to give developers direct access to enterprise-grade agentic capabilities from experimentation to production. Sovereign AI. For all. Download Command A : huggingface.co/CohereLabs/co… Or learn more: cohere.com/blog/command-a-pl…

Our first fully open source Apache 2 model :)

Cohere launches open weights model Command A that achieves 37 on the Artificial Analysis Intelligence Index The release of Command A places @Cohere in line with Claude 4.5 Haiku on the Intelligence Index, and just above NVIDIA Nemotron 3 Super and Gemini 3.1 Flash-Lite. Key Takeaways: ➤ Command A ranks first on AA-Omniscience Non-Hallucination at 86%, ~3 percentage points ahead of the next-best model. Its AA-Omniscience Accuracy is 9%, so the headline AA-Omniscience score lands at -4, demonstrating a similar archetype to Claude 4.5 Haiku, where the model knows its limits ➤ On Cohere’s API, Command A (~281 output tokens per second) is faster than several comparable open-weights and small to mid-sized proprietary models (e.g., GPT-5.4 nano, Claude 4.5 Haiku, and Grok 4.3), but still slower than Gemini 3.1 Flash-Lite Preview, which outputs 304 tokens per second ➤ Command A trails its peer set on scientific reasoning (HLE ~11%, GPQA Diamond ~76%) and on coding (Terminal-Bench Hard ~25%, SciCode ~38%), consistent with gaps on the hardest science and agentic coding benchmarks ➤ It supports visual reasoning and scores 63% on MMMU-Pro (between Claude 4.5 Haiku at 59% and GPT-5.4 nano (xhigh) at 65%)

Introducing: Cohere Command A We’ve created our most powerful LLM yet, optimized it to run on as little hardware as possible, and released it open-source for all.

🚨 Do LLMs need to store everything they read in memory? To reduce KV cache size and improve decoding speeds, we propose Self-Pruned KV attention, a mechanism where the model learns to decide which KVs to write in the persistent KV cache, discarding all the rest! @AIatMeta🧵

@JinaAI_ has hopped on the omnimodal train🚂 They just dropped a collection of two Omni embedding models (0.9B & 2B). Similar to BidirLM, they seem to rely on the Qwen modality head for the larger one, while sticking with EuroBERT for the nano version 🥰 huggingface.co/collections/j…

BidirLM-Omni is on MTEB and Sentence-Transformer! huggingface.co/spaces/mteb/l… 🥇#1 Open-Source Model on MTEB (#15 overall) 🖼️#1 across all sizes on MIEB (Image) 🎧#1 sub-7B model on MAEB (Audio, #2 overall) Small size, massive performance, Fully open Model: huggingface.co/BidirLM

We are currently presenting 'Should We Still Pretrain Encoders with Masked Language Modeling?' Come see us in Hall 3 #1304 @iclr_conf arxiv.org/abs/2507.00994

If you're at ICLR, come say hi to @orionweller and @N1colAIs! And also, shot-out to all the French people pushing out the encoder architecture, it seems like, as for ColBERT, French taste is unmatched! (Non exhaustive list, pardon me but Twitter search is bad): @gisship @PierreColombo6 @ManuelFaysse @pteiletche @MaceQuent1 @mlpc123

Great work from @gisship and @N1colAIs from @centralesupelec

BERT-as-a-Judge A robust alternative to rigid lexical matching for LLM evaluation. Matches the performance of LLM-as-a-Judge at a fraction of the computational cost.

Encoders are so much better for classification, why not use them for judging? Awesome study from @N1colAIs - cool to see a 210m BERT model beating much larger Qwen and Gemma models.

Evaluation is underrated. If your eval signal is noisy, you're flying blind. BERT-as-a-Judge gives you a fast, cheap way to improve your signal-to-noise ratio without spinning up a full LLM judge. Exactly the kind of infra work that compounds. @gisship @N1colAIs congrats!

🎉 Second paper this month! Introducing BERT-as-a-Judge (x @gisship) ⚖️ Evaluating LLMs with rigid lexical methods often fails right answers due to bad formatting. While "LLM-as-a-Judge" solves this, it remains costly & slow. Our fix? A lightweight, encoder-driven approach.

There's a wave of omni embedding models (gemini, nemotron, bidirlm). Excited to support this trend with our multimodal mteb versions (mieb, maeb) - video coming soon🎥

Omni embeddings are becoming the new standard. Glad to see @N1colAIs @Muennighoff pushing multimodal eval forward with MIEB & MAEB — can't wait for the video!

They even released the base bidirectional models 😍 Great release, thanks for all the checkpoints ♥️