Together AI

Reserved GPU cluster prices increased across all duration tiers: NVIDIA HGX H100 from $2.69-$2.25/hr to $2.99-$2.55/hr; H200 from $3.19-$2.59/hr to $3.49-$2.89/hr; B200 from $5.49-$4.49/hr to $7.15-$6.39/hr (30% increase). Three models removed from Together AI serverless inference pricing list: Llama 4 Maverick ($0.27/$0.85/1M tokens), Qwen3-Next-80B-A3B-Instruct ($0.15/$1.50/1M tokens), and GLM-4.5-Air ($0.20/$1.10/1M tokens).

Three LLM model prices removed: Qwen3 235B A22B Thinking output ($0.65), Sora 2 Pro ($2.40), and Kimi K2 Thinking output ($4.00)

Dedicated GPU instance prices increased: H100 80GB $3.36->$3.99/hr, H200 141GB $4.99->$5.49/hr. GPU Clusters on-demand hourly also up: H100 $2.99->$3.49, H200 $3.79->$4.19, B200 $5.50->$7.49. Product sections renamed: Dedicated Endpoints is now Dedicated Inference; GPU Cloud is now GPU Clusters with restructured reservation tiers (1W-1M, 2-3M, 4-6M, 6+ months).

Quarterly roll-up · includes 7 weeklies

MiniMax M2.5 now shows a $0.06 (cached) input price alongside $0.30 input / $1.20 output Serverless Inference model lineup updated: Qwen3.5 9B added ($0.10/$0.15); Llama 3.1 405B, Ministral 3 14B, Llama 3.1 8B, GLM-4.6, and Llama 3.2 3B Instruct Turbo removed

Llama 4 Maverick OUTPUT price introduced at $0.85; Dedicated Inference pricing restructured [H100 80GB increased $3.36→$3.99; H200 141GB increased $4.99→$5.49; B200 180GB added at $9.95; GPU Clusters hourly rates increased: H100 $2.99→$3.49, H200 $3.79→$4.19, B200 $5.50→$7.49; GPU Clusters 1-week rates changed: H100 $2.50→$2.69, H200 $3.45→$3.19, B200 $4.90→$5.49; Reserved pricing periods restructured from 3 time bands to 5 bands] Removed model options from Serverless Inference [Llama 3 70B Instruct Reference removed; Llama 3.1 70B removed; DeepSeek-R1-0528 Throughput removed; Qwen3 235B A22B FP8 Throughput removed; Qwen2.5 72B removed]

Removed M2-BERT 80M 32K Retrieval ($0.01/1M tokens) embedding model and Salesforce LlamaRank ($0.10/1M tokens) reranking model from catalog. Added 3 new text models: MiniMax M2.5 ($0.30/$1.20 per 1M tokens), Qwen3.5-397B-A17B ($0.60/$3.60 per 1M tokens), and GLM-5 ($1.00/$3.20 per 1M tokens) to Serverless Inference catalog.

DeepSeek-V3.1 output price increased $1.25→$1.70/M tokens; Mistral Small 3 prices decreased $0.80/$0.80→$0.10/$0.30/M tokens.

FLUX.1 Krea [dev] image model price increased from $0.60 to $0.88 per MP; DeepSeek-V3-0324 pricing restructured from $0.55 input / $2.19 output to flat $1.25 per 1M tokens; Qwen3-235B-A22B Instruct 2507 FP8 pricing reduced from $0.65 input / $3.00 output to $0.20 input / $0.60 output per 1M tokens (69% decrease); Kimi K2 Thinking model pricing reduced from $1.20 input / $4.00 output to $0.60 input / $2.20 output per 1M tokens (50% decrease); MiniMax Hailuo 02 video model pricing increased from $0.28 to $0.49 per video (75% increase); Google Veo 3.0 pricing increased from $2.50 to $3.20 per video (28% increase); Google Veo 3.0+ Audio pricing decreased from $1.60 to $1.20 per video (25% decrease)

NVIDIA HGX B200 Instant Cluster GPU prices increased: 1-week commitment $4.00→$4.50/hr per GPU, 1-6 day commitment $4.50→$4.90/hr per GPU. Hourly rate ($5.50) unchanged. DeepSeek-V3.1 output inference price decreased from $1.70 to $1.25 per 1M tokens on the Together AI serverless inference platform.