Can Claude Make Music?

This is the fourth time we've asked "can this chatbot make music?" — after Grok, ChatGPT, and Gemini — and Claude turns out to be the strangest case of the four. The headline answer is the same as ChatGPT's: no audio. But the details underneath it are different enough, and interesting enough, that copying our ChatGPT verdict over would miss what Claude actually is. We build an AI music app, we use these assistants daily, and yes — we're aware of the comedy of fact-checking an article about Claude partly by asking Claude. We verified against what Anthropic has actually shipped instead.

The 60-second answer

• Claude produces no audio. Full stop. No vocals, no instruments, no sound files. Anthropic has no music model — no equivalent of Google's Lyria, xAI's Grok Imagine, or OpenAI's Sora — and none has been announced. Claude's voice mode talks; it doesn't sing.
• As a lyricist, it's arguably the best of the chatbots. That's our team's testing opinion, and we'll defend it below.
• The asterisk: Claude can make music as code. MIDI files, notation, browser-synth sketches. The sound then comes from whatever runs the code — an instrument, a plugin, a browser — not from Claude.
• The 2026 twist: Claude can operate music tools. Through tool connectors, Claude can send prompts to external music generators and hand you back results. Conductor, not singer.
• For a finished song, the answer is unchanged: a dedicated AI music app — Sonx, Suno, Udio — is where the actual track comes from.

Claude is the session writer who can also read the mixing desk manual. What it still can't do is make a single sound.

Claude the lyricist

Every major chatbot writes lyrics. In our internal testing, Claude's are the ones that most often survive contact with a real song. The difference is restraint: where other models reach for the biggest rhyme and the third repetition of the hook, Claude is better at the unglamorous parts of lyric writing — lines that scan naturally when sung, images that are specific instead of grand, verses that leave room for the melody to do work. Ask the models for "a sad song that doesn't say the word sad" and you'll hear the difference quickly.

It's not magic, and it still needs an editor — every AI lyric does. The same rules we gave for ChatGPT apply here: be specific about mood, genre, story, and structure; push back on the first draft; cut anything that sounds like it was written by a committee of greeting cards. And Claude is just as good at the sleeper job we keep banging on about: turning a vague idea into a detailed music prompt. Our study of 1,136 real AI music prompts found the median prompt is seven words long — a chatbot that expands those seven words into tempo, instrumentation, and vocal direction fixes the single most common AI-music mistake before you've generated anything.

The asterisk: music as code

Here's where Claude genuinely differs from the rest of this series. Claude has a reputation as the strongest coding assistant of the major models, and music, it turns out, is something you can code.

Ask Claude for a chord progression and it can hand you a MIDI file — actual note data you can drop into GarageBand, Logic, or any DAW and hear played by real instruments. There's a small ecosystem built on exactly this: plugins that sit inside your DAW and use Claude to generate editable melodies, chords, basslines, and drum patterns as MIDI. Ask it for a browser synth sketch and it writes a small web page that actually plays a loop out loud. Ask for notation, and you get something a musician can read.

Be clear about what this is and isn't. MIDI is instructions, not sound — the audio comes from whatever synth or sampler plays it. Nothing in this pipeline sings, and nothing produces a finished, mixed track. But if you're a producer who wants raw material to shape yourself, "Claude, give me a melancholy chord loop in F minor as MIDI" is a legitimately useful trick that Grok, ChatGPT, and Gemini users mostly don't get in this form.

Claude the operator

The other 2026-specific wrinkle: Claude has become the assistant most associated with using tools — connectors that let it operate other software on your behalf. Music services are on that list. Wire a music generator up to Claude and you can type "make me a dreamy synth-pop track about leaving the city" into a chat; Claude refines the idea, sends it to the connected service, and hands back the result.

It's a glimpse of a real future — assistants as producers who hire the right specialist for each job. But notice who's doing the singing: not Claude. The audio still comes from a dedicated music model at the end of the chain, and the connector route adds setup friction that only makes sense if you're already living in Claude all day. If what you want is simply a song, opening a music app remains the shortest path between the idea and the sound — the same conclusion this series keeps arriving at, because it keeps being true.

The Claude + Sonx workflow

1. Draft lyrics in Claude. Mood, genre, story, structure — and ask for restraint: "specific images, no clichés, lines that scan when sung."
2. Edit ruthlessly. Claude's drafts need less surgery than most, but the song still has to sound like you.
3. Have Claude write the music prompt too. Tempo, instrumentation, vocal style, mood, one paragraph. This is the seven-word-prompt fix.
4. Paste both into Sonx. Lyrics in the lyrics field, description as the prompt, pick a genre, generate.
5. Iterate and export. Regenerate the chorus until it lands, add a music video if it's headed for social — our TikTok guide covers the hook-first structure — then export and post.

TL;DR

Claude can't make music — no audio, no singing voice mode, no Anthropic music model. What it can do is write the most usable lyrics of the major chatbots, generate music as code (MIDI and browser sketches whose sound comes from other software), and operate external music tools through connectors. Every road still ends at the same place: a dedicated AI music app turns the words into an actual track. Claude writes; something else has to sing.