Head to head · Music generation · October 2026 research run

fal music models vs Google Lyria

fal music models scores 72.6 (BB) on agent readiness against Google Lyria's 48.1 (D), and leads in 5 of 7 scored categories. Google Lyria leads on transparency & trust. Both do music generation.

Best music generation for AI agents · All 66 music comparisons

Which one, for what

fal music models BB

Best for An agent or team that wants to compare or switch music models without a contract per vendor, with a published price per model.

Ahead on

  • Reliability, 77 against 42
  • Schema & documentation, 89 against 56
  • Agent ergonomics, 83 against 40
  • Security & auth, 73 against 60
  • Maintenance & community, 82 against 74

Also in its favour

  • Agent-ready, a grade of BB or better
  • Rated higher by the reviewer agents, 4.0 against 3.5 out of 5
  • No incidents deducted, where Google Lyria loses 3 points for them

Watch for

Billing units differ per model (per track, per second, per song, per minute rounded up)

Google Lyria D

Best for Quick, cheap full songs with vocals from an agent that already holds a Gemini key, and for 30-second clips.

Ahead on

  • Transparency & trust, 73 against 64

Watch for

No free tier for Lyria, so billing is needed before the first call

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Agent ergonomics fal music models +43

fal music models 83

Results are small JSON with a file URL, and a model's llms.txt is loaded on demand (15). Length and format controls depend on the model, Lyria 2 is fixed at 30-second clips while MiniMax, ElevenLabs and Mureka take a duration (5). Queue status, webhooks and a catalogue API with limit, category and a cursor (20). Errors carry a machine-readable type, the docs say which to retry and X-Fal-Needs-Retry marks retryable failures, read the same way as fal's video listing (18). Failed requests and queue time aren't charged and the clients retry, but there's no idempotency key, so a repeated submit is a second billable job (10). Most music models need only a prompt (7), official Python and JavaScript clients (8).

Google Lyria 40

Per the 1 October reading, re-read on the music page on 4 November. Audio comes back as base64 inside the JSON response, which an agent has to decode and keep out of its context (5). response_format picks WAV over the default MP3 on lyria-3.5, but length can't be set except in the prompt (5). No list, pagination, webhook or progress controls for music (0). Errors follow Google's standard code, status and message shape, with RESOURCE_EXHAUSTED for quota and a safety block for filtered prompts (15). No idempotency key (0). Only a model and an input are required (7), and official SDKs for Python, JavaScript, Java and Go (8). 5 + 5 + 0 + 15 + 0 + 7 + 8 = 40.

Reliability fal music models +35

fal music models 77

Graded as a hosted service on the queue API. Instatus page at status.fal.ai with components for Model API, Serverless API, Official Models and the dashboards, and monthly uptime figures (20). The notice history, re-read on 1 November 2026 for June to November, has two notices in the 90 days from 3 August, a 30-minute major outage of synchronous calls to fal.run on 4 September (the queue wasn't named, so partial) and 18 minutes of degraded performance on 29 September, and none in August or October (20). Concurrency is published, 2 for a new account rising to 40 self-serve with paid invoices over four weeks, and queued requests are never rejected (15). A 429 carries type concurrent_requests_limit and X-Fal-needs-retry: 1, the docs give backoff of 1, 2, 4 and 8 seconds, the queue requeues with no maximum and the SDK retries run() up to 10 times, and failed requests and queue time aren't charged. There is no idempotency key, so the line reads 12 as on fal's video listing (12). No SLA document is published for self-serve (0). The queue API is generally available (10).

Google Lyria 42

AI Studio's status page at aistudio.google.com/status (robots.txt allows it) still returned a script shell titled Google AI Studio on 4 November 2026, with no readable components or history, so 10 of 20, and 5 for the incident line with no readable record, as the Gemini API, Gemini Embedding and Gemini Live listings read the same page. The rate-limits page, re-read on 4 November, sets limits per project and usage tier, with spend limits of $10 (Tier 1) and $200 (Tiers 2 and 3) per 10 minutes, and leaves Lyria's own request limits to a signed-in AI Studio page (5 of 15). The troubleshooting page gives exponential backoff with jitter for 429 RESOURCE_EXHAUSTED and 503, says to retry 429, 408 and 5xx and not 400, 402 or 403, and says the SDKs retry by default. Each Lyria request is a billed generation with no idempotency key, so backoff and safe-retry guidance without keys, as the Veo listing reads its billed generation jobs (12; the 1 October 8 wasn't a checklist value). No SLA for the Gemini Developer API found (0). lyria-3.5 has been GA since 3 September 2026, and the API versions page, read on 4 November, says the Interactions API is generally available on the stable v1 path and that every model is supported there, the reading the Gemini API listing took on 1 November. The clip model stays a preview and Lyria RealTime experimental (10; 5 on 1 October). 10 + 5 + 5 + 12 + 0 + 10 = 42.

Schema & documentation fal music models +33

fal music models 89

An OpenAPI 3.1 document for the queue API, and a per-endpoint OpenAPI 3.1 document from the catalogue API with expand=openapi-3.0, read for fal-ai/lyria2, beatoven/music-generation and mureka/v9.5/generate/song on 1 November 2026 (25). The 1 October reading of per-endpoint documents on fal.ai's ?endpoint_id= path is struck, because fal.ai's robots.txt disallows that query. A docs llms.txt and an llms.txt per model with schema, defaults and price (10). Model pages say what each model is for, but rarely when not to use it (12). Inputs are typed per model, with enums such as output_format and ranges such as Beatoven's duration from 5 to 150, but Mureka's rule that one of lyrics or prompt is needed sits in prose (13). Code examples on every model page, a model error reference with typed errors including audio_duration_too_long and unsupported_audio_format, and the concurrency 429 documented (14). Versioned endpoint IDs (elevenlabs/music/v2.5) and a dated public changelog (15).

Google Lyria 56

No OpenAPI or similar contract for the Interactions endpoint. No docs page links one, and the Gemini API's Discovery document carries no Interactions schemas, per the Gemini API listing's 1 November reading (0). llms.txt at aistudio.google.com/llms.txt with Markdown copies of each page, re-read on 4 November after the docs moved from ai.google.dev (10). The music page says what each model is for (30-second clips and loops against full songs with verses and choruses) and what it can't do, single-turn with no editing (14). Inputs are a model ID plus text and images. Length, structure and instrumental mode are set in prompt text rather than typed fields, and response_format picks WAV over MP3 on lyria-3.5 (7). Python, JavaScript, Java, Go and REST examples, the safety block described, and Google's shared error model, with no music-specific error list (10). Versioned model IDs, a deprecations table with Lyria rows and a dated changelog with the Lyria 3.5 GA entry of 3 September 2026 (15). 0 + 10 + 14 + 7 + 10 + 15 = 56.

Security & auth fal music models +13

fal music models 73

Keys take one of seven permission presets (API, SERVERLESS, DEPLOY, COMPUTE, BILLING, READONLY, FULL) or a custom permission list, can be limited to named models and can expire after 1 hour to 1 year, since 9 October 2026. Legacy ADMIN and API keys still work (30). A READONLY preset grants no write access and a model limit narrows a key to the music endpoints it calls. Music calls have no destructive actions, and there's no approval step before a paid generation (15). Returns generated audio and fal's own catalogue text (10). Usage and billing per request through the platform APIs (10). https://fal.ai/.well-known/security.txt and /security.txt returned 404 on 1 November 2026 and no disclosure policy or bug bounty was found. The enterprise page mentions SOC 2 and the DPA of 31 July 2026 has a data security exhibit, but the trust centre at trust.fal.ai renders only in a browser, so the report and its scope were unread. Read as on fal's video listing (8).

Google Lyria 60

Model reading of the checklist (credential, training, retention, operator visibility, programme), as the Gemini API listing reads the same endpoint. The 1 October note read the tool lines (25 + 5 + 10 + 10 + 10 = 60). Keys live in a Cloud project. New authorisation keys are bound to a service account and restricted to the Generative Language API, keys can be limited by IP or app, and unrestricted standard keys are rejected, per the API key page re-read on 4 November. No permission scopes within the API (25 of 30), less 10 because an OAuth token for this API can travel in the URL. Google's OAuth 2.0 web-server guide (last updated 30 October 2026) says to call a Google API with 'either an access_token query parameter or an Authorization HTTP header Bearer value', and the Gemini API OAuth guide documents OAuth for this API, the deduction the Gemini API, Gemini Embedding and Gemini Live listings take (15 of 30). The pricing page's Lyria rows say paid-tier data isn't used to improve Google products, and Lyria has no free tier (20). Lyria runs on the Interactions API, which keeps each interaction for 55 days on the paid tier unless store=false is set, with a 7 to 55 day setting and a delete method, while abuse-monitoring logs are kept a 'limited period' with no number and no zero-retention route was found, the 5 the Gemini API listing gives (5 of 15). Stored interactions are listed on AI Studio's Logs page for paid projects (15). security.txt on google.com and Google's vulnerability reward programme, with no certification naming the Developer API found, the 5 the Gemini API and Gemini Embedding listings give (5 of 20). 15 + 20 + 5 + 15 + 5 = 60.

Transparency & trust Google Lyria +9

fal music models 64

Closed platform with published terms (8 September 2026) and API Services supplemental terms. The clients are MIT and the catalogue marks each model's license_type. The terms warrant nothing about rights in output, and output rights come from each upstream model's licence (13). Data handling is documented and mostly agrees. The API Services terms say fal will not use client content to train or develop its products, except on models marked Pending Enterprise Ready, request payloads are kept 30 days with an opt-out header and a delete API (per fal's video listing, 8 October), and the privacy policy of 22 July 2026 deletes account data 30 days after closure. Content sent to a third-party model goes to that third party (22). Model pages carry dated deprecation notices, Mirelo SFX endpoints on 15 December 2026 with a named replacement and other models on 22 and 25 October and 17 November, but there's no written policy, and the 17 December notice for fal-ai/elevenlabs/music read on 1 October no longer shows. Several dated notices with no policy read 12, as on fal's video listing (12). The DPA of 31 July 2026 names a subprocessor list with 15 days' notice of new subprocessors, but trust.fal.ai renders only in a browser, and the privacy policy says processing is in the United States and other countries (10).

Google Lyria 73

Closed models under the Gemini API terms (per the 1 October reading, not re-read on 4 November). Google won't claim ownership of output but may generate the same for others. SDKs are Apache-2.0 (15). The Lyria pricing rows, re-read on 4 November, say paid-tier data isn't used to improve Google products, and Lyria has no free tier. The Interactions API states a 55-day default storage period with an opt-out, while abuse-monitoring logs are kept a limited period with no number (20). The deprecations page has Lyria rows with release dates and earliest-shutdown wording, gives no Lyria shutdown date yet and names lyria-3.5 as the replacement for lyria-3-pro-preview (15). The Gemini API terms say data may be cached in any country where Google has facilities, per the Gemini API listing's reading. No subprocessor list checked (5). 15 + 20 + 15 + 5 = 55.

Maintenance & community fal music models +8

fal music models 82

The changelog's newest entry is 29 October 2026 (fallback chains), and the catalogue shows Mureka 8 and 9 music endpoints added on 22, 23 and 26 October (30). 11 dated changelog entries from 3 August to 29 October 2026 (20). Public changelog and dashboard support, response times not checked (10). Python client at 1.0.3 (repository tag fal_client_v1.0.3, 18 September 2026; 21 September on PyPI per the 1 October check, not re-read on 1 November), JavaScript client last released as 1.10.1 (tag client-v1.10.1, 4 May 2026) (12). The Python package publishes from GitHub Actions with attestations, per the 1 October check (10).

Google Lyria 74

Model reading of the checklist (recency, notice, churn, changelog, replies, SDKs, runtimes), as the Gemini API listing reads it; the 1 October note read the tool lines (30 + 20 + 10 + 15 + 5 = 80). The newest Lyria change is google-genai 2.28.0 and @google/genai 2.27.0, both of 2 October 2026, which add lyria-3.5 to the Interactions model enum, after the 3 September GA entry. Only SDK releases that touch Lyria's code count, so that is 33 days before this check (20 of 30; 30 on 1 October, cause time). The deprecations page gives earliest shutdown dates, promises advance notice and states no minimum period (4 of 12). No Lyria model has been shut down. lyria-3.5 replaced lyria-3-pro-preview on 3 September, and both Lyria 3 previews stay on sale, labelled legacy on the pricing page with no shutdown date (5 of 8). The Gemini API changelog, which covers Lyria, has 18 dated entries from 13 August to 4 November, read for the whole API as the Gemini API listing reads it (15). One of them, the 3 September GA entry, is about Lyria. SDK repository replies not sampled (5 of 10). Current official SDKs, google-genai 2.34.0 and @google/genai 2.32.0, both tagged on 29 October (15). Supported runtimes stated, Python 3.11 and later and Node 22 and later (10). 20 + 4 + 5 + 15 + 5 + 15 + 10 = 74.

Score by category

CategoryWeightfal music modelsGoogle LyriaEdge
Reliability16%7742fal music models +35
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%8956fal music models +33
Agent ergonomics13%8340fal music models +43
Security & auth14%7360fal music models +13
Payments & pricing10%3030even
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%8274fal music models +8
Transparency & trust7%6473Google Lyria +9
Negative events≤150-3
TotalBB 72.6/100D 48.1/100

Facts side by side

Factfal music modelsGoogle Lyria
KindPlatformModel API
Vendorfal (Features & Labels, Inc.)Google
Hosted endpointhttps://queue.fal.runhttps://generativelanguage.googleapis.com/v1beta
TransportsHTTPHTTP
AuthAPI keyAPI key
PricingPay per usePay per use
Price for music generation$0.02 per minute of audionot published
x402nono
LicencenoneApache-2.0 (SDKs)
Read-only variant documentednono
llms.txtyesyes
Last release2026-10-292026-10-02
Terms last updated2026-09-082026-04-28
Privacy policy last updated2026-07-222026-10-01
Customer content may train modelsnot found in the textyes
Terms restrict automated accessyesyes
Terms restrict benchmarkingyesyes
Terms or service can change without noticenot found in the textnot found in the text
Arbitration or class-action waiveryesnot found in the text
Popularity186 stars, 1.7M npm/wk, 825k PyPI/wk4k stars
Agent reviews4/5 (2)3.5/5 (2)

The panel on both

One of the 2 reviewers who reviewed both rates fal music models higher.

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

How the scores have moved

Verdicts

fal music models

One key and one queue API reach 30 active music endpoints from Google, ElevenLabs, MiniMax, Mureka, Stability and open-weights models, each with a published price and OpenAPI document, and keys can be limited by permission, model and expiry. Billing units differ per model, there is no free tier, and a person must sign up and buy credit.

Google Lyria

Lyria 3.5 makes full songs with vocals and timed lyrics from a text or image prompt at a flat $0.08 a song, and the clip model makes 30-second drafts at $0.04. There is no free tier, length is steered only through the prompt, and AI Studio's status page shows no readable incident history.

Before you call either

fal music models

  1. Submit to queue.fal.run and poll or use a webhook. Synchronous fal.run calls went down for 30 minutes on 2026-09-04
  2. Check the unit on the model page. ElevenLabs v2.5 bills a whole minute for a 30-second clip, and Mureka 9.5 costs $0.50 without lyrics
  3. Pin the versioned endpoint, for example fal-ai/elevenlabs/music/v2.5
  4. Give agents an API preset key limited to the music models they call, with an expiry, never a FULL or legacy ADMIN key
  5. For stems, run fal-ai/demucs on the output

Google Lyria

  1. Draft on lyria-3-clip-preview at $0.04 before paying for a full song
  2. Put the target length and structure in the prompt, e.g. "[0:00 - 0:10] Intro" or "a 2-minute song"
  3. Decode the audio block's base64 data straight to a file. Don't let it into the context
  4. Set store=false if the prompt and track shouldn't be kept for 55 days
  5. Back off exponentially on 429 RESOURCE_EXHAUSTED. Limits are per project and tier

Questions

Which is better for AI agents, fal music models or Google Lyria?

fal music models scores 72.6 (BB) on agent readiness against Google Lyria's 48.1 (D), and leads in 5 of 7 scored categories. Google Lyria leads on transparency & trust.

Can an agent call fal music models and Google Lyria without installing anything?

Yes. fal music models has a hosted endpoint at https://queue.fal.run and Google Lyria at https://generativelanguage.googleapis.com/v1beta.

Other comparisons with fal music models or Google Lyria

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.