Together AI vs Replicate: FLUX image API costs

For a 1 MP FLUX.2 dev text-to-image output, Replicate charges $0.014 in standard mode (go_fast=false) or $0.012 in its default go_fast mode, below Together AI’s $0.0154+ per-image estimate, which states no output size; choose Replicate for that published rate, or Together for images alongside its LLM APIs.

Updated · 7 sources · Methodology

DimensionWinnerMarginEvidence
Price at 1 MP outputReplicate$0.0014+$0.0154+ / $0.014 (go_fast=false) · $0.012 (go_fast=true)127
Published generation timeNot ranked—<10 s / 5.2 s · vendor-reported · different workloads
OutputTie0Text → image · FLUX.2 dev3

Performance

MetricTogether AIReplicate
Generation configurationFLUX.2 dev · 32B3FLUX.2 dev · 32B3
Output resolution1 MP61 MP1
Published generation time<10 s · vendor-reported, 202555.2 s · one example, image edit with go_fast=true, 20251

Prices and request cost

ProviderPrice at 1 MP output1,000
Together AI27$0.0154+$15.40+
Replicate · go_fast=false1$0.014$14.00
Replicate · go_fast=true1$0.012$12.00
Cost example: 1,000 text-to-image outputs at 1 MP
Together AI$15.40+
Replicate · go_fast=false$14.00
Replicate · go_fast=true$12.00

Capabilities and serving conditions

DimensionTogether AIReplicate
Serving and capacityTogether lists FLUX.2 dev at an estimated $0.0154+ per image, passing through an external provider’s charge that varies with resolution. Image APIs share an account with LLM inference.27Replicate exposes the model as a prediction API. Public models can use output-based billing, while private deployments also bill setup and idle hardware time.1

When each option fits

Together AI

  • Choose Together when the same application already uses its LLM inference and should add images under the same service account.26
  • Its $0.0154+ figure is an estimate rather than a rate; Together advises generating one image at the target resolution and checking the charge before scaling up.27

Replicate

  • Choose Replicate for a published per-megapixel rate that is lower at 1 MP in both go_fast and standard mode, and for applications built around model prediction APIs.1
  • Reference-image inputs are separately billable by MP; private deployment billing also includes setup and idle hardware.14

Methodology and sources

Together’s <10 s is a FLUX.2 launch spec from 2025-11-25 with no stated resolution or workload. Replicate’s 5.2 s is the predict time of one published 2025-11-25 example that edits a reference image with go_fast=true at 816×1224. The figures describe different workloads, so they do not establish a speed margin.51

The price comparison is text-to-image only, with no reference input image and a 1 MP output. Replicate bills FLUX.2 dev per output megapixel: $0.012 with go_fast=true (its default) or $0.014 with go_fast=false. Together’s $0.0154+ is a per-image estimate of a pass-through provider charge that varies with resolution; Together does not state the output size behind it.

Image charges exclude tax, credits, retries and private deployment idle time. Input reference images, larger output dimensions and provider rounding can change the bill. Novita AI presents public endpoint prices, not a private image-quality or latency test.

  1. Replicate · FLUX.2 dev
  2. Together AI · image API pricing
  3. Black Forest Labs · FLUX.2 dev model card
  4. Replicate · public and private deployment billing
  5. Together AI · FLUX.2 launch and generation time
  6. Together AI · FLUX endpoint parameters
  7. Together AI · how image models bill

Build with Novita Model APIs

Novita AI doesn’t host FLUX. Its model catalog includes other image generation models alongside LLM APIs, under one API key.

Start BuildingExplore Models