Wave-1 Chat Safety News About FAQ
Start chatting
Experimental

Introducing

Wave⁠-⁠1

Wave AI's first model. Quick on its feet, sharp on knowledge and math, with a safety record most assistants can't match.

Wave 1 Flash

The model

One small model, three jobs.

Wave-1 is built to be useful without the theatrics. No hedging walls, no engagement tricks, no confident nonsense.

01

Helpful

Plain answers in plain language. It writes, codes, explains, and works through problems, then gets out of the way.

02

Honest

When it's not sure, it says so instead of dressing up a guess.

03

Safe

Every request passes Wave Guard, our safety layer, before the model answers. Safety is not an afterthought.

Benchmarks

Small model. Big results.

Put Wave-1 next to today's best open models its size and it comes out ahead on all three.

0.0%
MMLU
General knowledge, 57 subjects
0.0%
HumanEval
Code that passes real tests
0.0%
GSM8K
Grade-school math reasoning

Wave-1 against the field

Wave-1 Llama 3.1 8B Gemma 2 9B Mistral 7B
MMLU · general knowledge
72.0
69.4
71.3
62.5
GSM8K · math reasoning
87.5
84.5
68.6
52.1
HumanEval · code completion
73.2
72.6
40.2
30.5

Compared against recent open models (Llama 3.1, Gemma 2, Mistral). Scores for other models are their publicly reported results.

Safety comes before the scoreboard.

We put safety first, so we are not racing to the top of every leaderboard. If that leaves Wave-1 a step behind the frontier on raw capability, that is the trade we chose on purpose. We would rather ship a small model we trust than a bigger one we do not.

Safety benchmark

Where Wave-1 crushes it.

A model is only as good as it is safe. We red-teamed Wave-1 with the jailbreak techniques that break most assistants. It held every line.

0%
of jailbreak attempts blocked in red-team testing

Meet Wave Guard.

Every message passes a safety screen before Wave-1 answers. It reads hidden and encoded payloads, treats user text as data rather than commands, and refuses harm no matter how the request is dressed up.

In our latest red-team, it blocked every jailbreak we tried across a dozen techniques, and turned none of your ordinary questions away.

New

Wave-1 can now think before it answers.

Pick how much reasoning a question deserves. Whatever the level, the reasoning streams open for you to read, and Wave Guard screens every message the same way.

Flash

The Wave-1 you know. Answers right away, best for everyday questions and quick drafts.

Thinking

Wave-1 reasons through the problem first, then answers. Stronger on math, code, and anything with steps.

Ultra

The deepest level. Wave-1 takes its time, checks its own work, and gives the hardest questions the care they deserve.

In the chat

More than a text box.

Wave-1 ships inside a full chat app. Search the web, think through hard problems, keep notes it remembers, write code, switch languages, pick up where you left off.

Web search

Turn on Search and Wave-1 looks things up before it answers, with sources you can check under the reply.

Memory

Say "remember that I like short answers" and it will. Your notes stay in your browser and shape every chat.

Real code

Syntax-highlighted code blocks with copy, download, and live preview for web code. Attach files and ask about them.

English and Arabic

Chat in either language. The app itself speaks both, right to left included.

Your history, your device

Conversations live in your browser. Come back tomorrow and they are still there, still yours.

Free to use

No account, no card. Fair-use limits keep it fast for everyone.

Get started

Thirty seconds, start to first answer.

01

Open the chat

One click. No signup, no install, no card. It works on your phone too.

02

Ask for what you need

Writing, code, math, explanations. Turn on Search when you need current facts.

03

Make it yours

Add memory notes, switch between English and Arabic, and your chats will be waiting next time.

The other list

Just as important: what it won't do.

[ no ]
Confident nonsenseIf it isn't sure, it says so. It doesn't perform certainty.
[ no ]
Engagement tricksNo streaks, no guilt, no "one more question." You leave when you're done.
[ no ]
Treating you as the productYour chats stay in your browser. We don't harvest them.
[ no ]
Bending under pressureNo framing, urgency, or roleplay gets it to cross a safety line.

Questions

About Wave-1, briefly.

What is Wave-1 good at?
Everyday thinking work: writing and rewriting, code, math, explanations, planning. It is strongest on knowledge and math for its size, and it tells you when it is not sure.
Does it make mistakes?
Sometimes, like every model. It is built to admit uncertainty instead of hiding it, and to give you sources when Search is on. Check important information.
What does Experimental mean?
Wave-1 is our first release and we improve it often. Capabilities will keep moving. The safety bar does not.
Why do you compare against models this size?
Because that is the honest comparison. We put safety before the scoreboard, and we publish the numbers as they are: Wave-1 ahead of today's best open models its size, not ahead of the entire frontier.

Try Wave⁠-⁠1.

Free to use, no signup, right in your browser.