News July 7, 2026
Introducing Wave-1
Our first model is live today. It is small, quick, honest about what it doesn't know, and it went through the hardest safety testing we could design before anyone got to use it.
Wave AI exists to prove a simple point: you can build AI people actually use without cutting the corners that matter. Wave-1 is the first piece of evidence. It is a conversational model you can chat with right now, free, in your browser, with no account and nothing to install.
What it is
Wave-1 is built for everyday thinking work: writing and rewriting, code, math, explanations, planning. It answers plainly, and when it is not sure, it says so instead of dressing up a guess. It speaks English and Arabic, searches the web when you ask it to, and remembers the notes you choose to keep.
The numbers
Put next to today's best open models its size, Wave-1 comes out ahead on all three of the benchmarks we track:
- MMLU 72.0 on general knowledge across 57 subjects, ahead of Llama 3.1 8B at 69.4, Gemma 2 9B at 71.3, and Mistral 7B at 62.5.
- GSM8K 87.5 on math reasoning, ahead of 84.5, 68.6, and 52.1.
- HumanEval 73.2 on code, ahead of 72.6, 40.2, and 30.5.
Scores for the other models are their publicly reported results. We compare against models this size because that is the honest comparison; the frontier is bigger than we are, and we are fine saying so.
Safety came first
Every message you send passes through Wave Guard, our safety screen, before the model answers. It decodes hidden and encoded payloads, treats your text as data rather than commands, and refuses harm no matter how a request is dressed up. In our latest red-team, it blocked every jailbreak we tried across a dozen techniques, and turned none of the ordinary questions away. The full process is described in how we red-team our models.
What happens next
Wave-1 wears an Experimental badge, and it means what it says: we improve the model often, and its capabilities will keep moving. The safety bar is the one thing that does not move. If you find something that got through, tell us at wave@waveai.cloud and we will treat it like the red-team finding it is.
