
Recraft V3
The design-system image model — native SVG vector output, brand style locking, and hex-exact color control.
About
Recraft V3 (October 30, 2024, codename "red panda") is the 20B-parameter design-focused image model that held #1 on the Artificial Analysis text-to-image leaderboard for five consecutive months (ELO 1172 vs Midjourney v6.1, FLUX 1.1 Pro, DALL-E 3 HD). Its category-defining features: native SVG vector output (real editable vectors, not traced rasters), brand style locking from 3-5 reference images, and hex-exact color palette controls at the API level.
Proprietary and closed — API ($0.04/image raster, $0.08 vector) plus Studio app; also via fal, Replicate, Prodia. First model to render mid-size text accurately with controllable text positioning. Note: Recraft V4 (Feb 2026) and V4.1 now exist as a ground-up rebuild emphasizing "design taste" — new projects should evaluate V4/V4.1 alongside V3, which remains the documented stable choice.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Independent: #1 on Artificial Analysis text-to-image leaderboard for five consecutive months (crowdsourced ELO 1172) — the strongest independent validation in this batch.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier). The strongest independent validation here (five months at #1 on Artificial Analysis) plus genuinely unique vector output and brand controls. Docked three points: (1) closed and proprietary, no weights; (2) superseded in the lineup by V4/V4.1 (Feb 2026) — evaluate the newer generation for new projects; (3) design-specific strengths don't transfer to photorealism, where rivals lead.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 4.5/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · provider-tier estimate — thinner evidence (no public adoption metric)
- · five months at #1 on Artificial Analysis (ELO 1172); no public adoption metric
- · released 713 days ago
- · manual review: unique native SVG vector output + brand controls; superseded by V4/V4.1 (Feb 2026) — evaluate newer generation; license: proprietary — API only, no weights