Nick Frosst initially said Cohere's Command A+ was out and Apache 2.0; later corrected us on X
Frosst's first X post called Command A+ Cohere's best model and described it as "out now" and "open source Apache 2.0." After publication, he corrected us on X; availability and licensing remain unconfirmed.
By Ryan Merket · · updated
Why it matters
A permissive Apache 2.0 release from a major model builder could give startups and enterprises a viable, lower-friction option to deploy, adapt, and ship on their own terms.
An initial post on X from Nick Frosst (Nick Frosst (@nickfrosst)) described Cohere's newest model, Command A+, as "out now" and "open source apache 2.0" in a short announcement on X. After this article published, Frosst corrected us in a follow-up post on X. Pending Cohere's formal materials, treat the availability and licensing details as unconfirmed.
What Frosst first said
The original post is light on technical detail and links. It includes a short video clip but no docs, benchmarks, or repo in the text of the post. The initial language pointed to two signals that matter for builders: 1) this is a new Command-series model from Cohere, and 2) Frosst described it as available under Apache 2.0. Given his subsequent correction, those specifics should be considered provisional.
Why the Apache 2.0 detail would stand out
Apache 2.0 is a permissive license that generally allows commercial use, modification, and redistribution with attribution and a copy of the license. For startups and enterprise teams, that usually translates into fewer licensing surprises, the ability to vendor a model inside your stack, and room to fork or fine-tune without negotiating separate terms.
If Command A+ is ultimately released with weights and tooling under Apache 2.0, teams could potentially:
- Self-host in their own cloud or on-prem for data control.
- Tune or adapt for domain-specific use cases without additional license hoops.
- Fork and harden for compliance or performance needs while staying inside a permissive legal envelope.
What we do not know yet
Neither post includes key model specifics. Operators and researchers will be looking for:
- Access details: weights, checkpoints, and tokenizer availability; where to download.
- Inference profile: memory footprint, throughput on common GPUs, quantization options.
- Capabilities: context window length, tool use or function calling support, multilingual performance.
- Benchmarks: evals on standard suites and how they were run.
- Safety and usage: guardrails, allowed-use guidance, and any red-teaming notes.
- Tuning: supported fine-tuning paths and license scope for derivatives.
How founders can act amid uncertainty
Even before full docs land, potential licensing direction can inform planning. If procurement or compliance blockers constrain model usage, a permissively licensed release would smooth the path to pilot or production. Prepare a quick technical spike to test:
- Fit: does the model meet latency and cost targets on your target hardware.
- Quality: evaluate on your own tasks, not just public leaderboards.
- Integrations: confirm it drops into your existing inference stack or serving framework.
- Data posture: map how self-hosting vs. hosted access affects privacy requirements.
Frosst's line that it is Cohere's "best model yet" will raise expectations. The next pieces that matter are the receipts: a repo or model card, solid evals, and clear guidance on deployment. RuntimeWire will update as Cohere posts more concrete details or formal release materials.