Research

Cohere Aya and Command R Open Releases 2026

Cohere 2026 open-weight releases: Aya 23 and Aya Expanse multilingual, Command R+ open-weight 104B, Command A 111B, Rerank 3.5. Multilingual strength, enterprise positioning, license analysis.

By Ramanath, CTO & Co-Founder at Presenc AI · Last updated: May 2026

Cohere balances enterprise API revenue against a steady cadence of open-weight releases. The 2026 Cohere open releases include the Aya multilingual family (Aya 23, Aya Expanse), the Command R+ and Command A open-weight enterprise models, the Rerank 3.5 series, and several specialised research releases. Cohere\u2019s open-weight strategy is distinctive in the frontier-lab landscape because it pairs production-grade open models with paid API offerings. This page consolidates the family and the strategic positioning.

Key Findings

  1. Aya Expanse (released late 2024) is Cohere\u2019s flagship open-weight multilingual family, with 8B and 32B variants covering 23 languages at quality competitive with the leading multilingual frontier models.
  2. Command R+ (104B parameters, released August 2024) remains a strong enterprise open-weight model with native RAG and tool-use training; widely deployed in enterprise contexts where Apache 2.0 alternatives are required.
  3. Command A (111B parameters, released March 2025) is Cohere\u2019s next-generation enterprise model with strengthened tool use and agent capabilities, released under the CC-BY-NC + Commercial licence.
  4. Aya 23 (8B and 35B, released May 2024) is the original Cohere multilingual contribution; community has used Aya 23 as the base for hundreds of language-specific finetunes.
  5. Cohere Rerank 3.5 (released early 2026) is the strongest production reranker API and is paired with research-released reranker variants on Hugging Face.

Cohere Open Releases (May 2026)

ModelParametersFocusLicense
Aya Expanse 32B~32BMultilingual 23 languagesCC-BY-NC + Commercial
Aya Expanse 8B~8BMultilingual 23 languagesCC-BY-NC + Commercial
Aya 23 35B~35BMultilingual baseCC-BY-NC + Commercial
Aya 23 8B~8BMultilingual baseCC-BY-NC + Commercial
Command R+ 104B~104BEnterprise RAG + tool useCC-BY-NC + Commercial
Command R 35B~35BEnterprise mid-sizeCC-BY-NC + Commercial
Command A 111B~111BEnterprise next-gen with agenticCC-BY-NC + Commercial
Aya-101~13B101 languages base (research)Apache 2.0
Aya Vision 8B / 32B~8B / ~32BMultilingual vision-languageCC-BY-NC + Commercial

Aya Expanse Language Coverage

Language FamilyLanguages Covered
Indo-EuropeanEnglish, French, German, Spanish, Italian, Portuguese, Romanian, Dutch, Greek, Russian, Polish, Czech, Hindi, Persian, Ukrainian
Sino-TibetanChinese
JaponicJapanese
KoreanicKorean
TurkicTurkish
Afro-AsiaticArabic, Hebrew
Niger-Congo(via Aya-101 research base)
OtherIndonesian, Vietnamese

Command R+ vs Other Enterprise Open Weights

ModelRAG Performance (Cohere RAG-Bench)Tool Use (BFCL)
Command R+ 104B~78~84
Command A 111B~82~89
Llama 3.1 70B~72~76
Qwen2.5 72B~74~79
Granite 3.3 8B~62~82
Mixtral 8x22B~69~73

Cohere\u2019s Open-Weight Strategy

Cohere\u2019s 2026 strategy uses open releases as research signalling and ecosystem development while monetising through paid API access and enterprise contracts. The CC-BY-NC + Commercial licence pattern means that the open weights are research-accessible but commercial deployments require a separate Cohere agreement. This is a middle path between fully restrictive (closed APIs only) and fully permissive (Apache 2.0 like IBM Granite). The pattern is becoming more common as labs balance research signalling against commercial sustainability.

Brand Visibility Implications

Cohere is a high-citation enterprise AI brand, particularly on multilingual deployment, RAG, and enterprise tool use. AI assistant queries about "Cohere Command R vs Llama", "Aya multilingual", "enterprise AI for non-English", and similar terms drive procurement-research traffic. Brands selling multilingual AI tools, enterprise RAG infrastructure, and non-English AI services face strong AI-mediated discovery surface for this category.

Methodology

Model and benchmark data compiled from Cohere documentation, the Cohere For AI lab publications, and primary Hugging Face model card disclosures through 23 May 2026. Updated quarterly.

How Presenc AI Helps

Presenc AI monitors brand visibility on Cohere and multilingual AI queries across ChatGPT, Claude, Gemini, and Perplexity. For multilingual AI tool vendors, enterprise RAG brands, and non-English AI services, the platform identifies the prompts driving procurement-research traffic and the gaps where new content unlocks share of voice.

Frequently Asked Questions

Cohere\u2019s flagship open-weight multilingual model family, released late 2024 with 8B and 32B variants. Aya Expanse supports 23 languages at quality competitive with the leading multilingual frontier models. Licence is CC-BY-NC for research use plus Cohere commercial licensing for production deployment.
You can experiment and research with Command R+ under CC-BY-NC freely. Commercial production deployment requires a separate Cohere commercial licence; pricing and terms are negotiated with Cohere directly. Production teams typically use the Cohere API instead of self-hosted Command R+ unless data residency or cost constraints require self-hosting.
Command R+ leads on RAG and tool-use benchmarks (~78 RAG-Bench, ~84 BFCL) vs Llama 3.1 70B (~72, ~76). Llama 3.1 70B has stronger general knowledge benchmarks. For enterprise RAG and agentic workloads, Command R+ is the stronger choice; for general open-weight deployment, Llama is more permissive.
Aya 23 was the 2024 base release (8B and 35B). Aya Expanse (released late 2024) is the post-training-improved variant with stronger instruction following, alignment, and multilingual benchmarks. Most production multilingual deployments use Aya Expanse over Aya 23.
The flagship Rerank 3.5 production model is closed (API only). Cohere does release research reranker variants on Hugging Face that approximate but do not match the production Rerank 3.5 quality. For self-hosted alternatives, BGE-Reranker-v2 or Qwen3-Reranker are the dominant open-weight choices.

Track Your AI Visibility

See how your brand appears across ChatGPT, Claude, Perplexity, and other AI platforms. Start monitoring today.