FLUX.2
Black Forest Labs' Open-Weight Image Frontier
Black Forest Labs' November 2025 image model — five variants spanning a fully-open Apache 2.0 4B/9B "klein" tier up to a 32B pro flagship. FLUX.2 leads open-weight image generation on prompt adherence and text rendering, and is the de facto base for fine- tuning communities that previously used Stable Diffusion.
Cost
Free
Open weights — self-host
How are Intelligence, Speed & Cost bucketed?
Intelligence and Speed buckets are percentile ranks on
Artificial Analysis. Cost buckets are fixed dollar
thresholds keyed off output-token price ($/M out).
Intelligence
- Top 1%≤ 1%
- Top 5%≤ 5%
- Top 10%≤ 10%
- Good≤ 25%
- Medium≤ 50%
- Below avg> 50%
Speed
- Top 1%≥ 345 tok/s
- Top 5%≥ 237 tok/s
- Top 10%≥ 196 tok/s
- Good≥ 146 tok/s
- Medium≥ 90 tok/s
- Slow< 90 tok/s
Cost
- Freeopen weights · self-host
- Low< $1 / M out
- Moderate$1–5 / M out
- High≥ $5 / M out
Why it matters
Made open-weight image generation a credible competitor to closed services on technical quality (text rendering, prompt adherence). Continues the post-Stable-Diffusion lineage of fully-open foundation models in the visual domain.
Core Capabilities
Generative
Produces images, video, audio, or other media.
Multimodal
Combines text, vision, and audio in one model.
Vision
Understands images, scenes, and visual context.
Context Window
Context window not disclosed.
Availability
API
Available
Product / App
Available
Open Source
Not released
Enterprise
Contact sales
Pricing Model
Free / self-host
Open weights — pay only for compute.
Self-host Capability / Performance
Where this model sits relative to the middle 60% of models in the tree. All scores are 0–10 (higher is better).
Lower 20% Upper 80% This model
Lower 20% 20th percentile — 20% of models score below this This model Where the current model lands Upper 80% 80th percentile — only 20% of models score above this
Percentile boundaries are computed across every model in the tree that reports the underlying benchmark for each capability.
What it feels like
- Flux.2 Pro: 32B params, latent flow matching with a Mistral-3 24B vision-language backbone — not a stock diffusion model
- Open-weight win rates: 66.6% on text-to-image, 59.8% single-reference editing, 63.6% multi-reference editing
- Up to 4 megapixel output with reliable text rendering, color matching, and character identity across outputs
- Up to 10 reference images for character / brand consistency — actual product feature, not just a tech demo
- Hands, faces, fabrics, and small text are noticeably more accurate than competitor open models
- Five variants (Pro / Flex / Dev / Klein / VAE) cover hosted-only, adjustable, open-weight, and tiny-open use cases
Best use cases
- Brand and product visualisation requiring consistency across multiple outputs
- Self-hosted image pipelines where open weights matter (Dev variant)
- Replacing Midjourney for workflows that need text rendering or precise reference editing
- Edge / on-device generation via the Klein open-source tier (when shipped)
Tools to try
Not ideal for
- Pure photorealism leaderboards — Midjourney v7 and Imagen 4 still lead in stylised aesthetics
- Video generation (FLUX.2 is image-only)
- Latency-sensitive interactive editing — flow-matching at 4MP is not real-time