P0pricing
Together AI Updates Fine-Tuning and Inference Pricing, Introduces New Models and Post-Training Service
Together AI has significantly revised its fine-tuning and inference pricing, introducing new models and discontinuing several others. The company also launched a new 'Post-Training' service with its own pricing structure.
Before · Sep 2, 2026
LFM2.5-8B-A1B $0.12, FLUX.1 [schnell] $0.0027, HiDream-I1-Full $0.009, Orpheus TTS $15, Cartesia Sonic-2, Google Veo 3.0 $1.60, Vidu 2.0, Wan 2.2 I2V $0.31, Wan 2.2 T2V $0.66, Google Veo 3.0 Fast + Audio, Google Veo 3.0 Fast, Google Veo 3.0 + Audio $3.20. Fine-tuning: Up to 16B LoRA $0.48, Full Fine-Tuning $1.35. DeepSeek-V4 Flash $25.00, GLM-5 $40.
After
DeepSeek V4.1 Flash $0.006 (cached), Muse Spark Open $0.15 (cached) $4.25, Tev1 4B Experimental Free, Reflection Beam $0.40, NVIDIA Nemotron 3.5 Super VL $0.02 (cached) $4.50. Post-Training: On-demand training infrastructure, billed per minute for each active model resource. Price per GPU PROMO $11.00. Fine-tuning: Qwen3.5 0.8B LoRA $0.34, Full fine-tuning $0.84. Llama 4 Maverick $5.60, $14.00. Kimi K2.7 Code $6.25.
Why it matters. Competitors should care about the updated pricing for fine-tuning and inference, as well as the introduction of new models and the 'Post-Training' service, which could impact their own competitive offerings and strategies. The changes reflect a shift in Together AI's model availability and cost structure.