Together AI

New serverless model Tev1 4B Experimental added to model pricing table at $0.04 input with a Free output tier.

Qwen3.7-Max serverless inference pricing increased 25% — input from $2.00 to $2.50 per 1M tokens, output from $6.00 to $7.50 per 1M tokens (cached input unchanged at $0.25/1M tokens)

New model DeepSeek V4.1 Flash added to the serverless model catalog at $0.30/1M input tokens, $1.20/1M output tokens, and $0.006/1M cached tokens Qwen3.7-Max price increased from $1.25/1M input ($0.13 cached) and $3.75/1M output to $2.00/1M input ($0.25 cached) and $6.00/1M output

Three new models added to Serverless Inference Chat pricing: Qwen3.8-2.4T-A95B ($2.50/$6.25, $0.50 cached), Muse Glimmer 30B ($0.35/$1.50, $0.04 cached), DeepSeek V4 Pro 0813 ($1.32/$3.96, $0.13 cached) Provisioned Throughput per-PTU capacity increased: MiniMax M3 TPM/PTU rose from 138,840/694,200/23,140 to 166,667/833,333/41,667; GLM-5.2 output TPM/PTU rose 9,620 to 11,364. Price unchanged, $0.05/PTU/min.

NVIDIA HGX H100 GPU cluster reserved pricing increased across all commitment tiers: 31-90 days $3.59→$3.69/GPU/hr, 91-180 days $3.29→$3.45/GPU/hr, 181+ days $3.09→$3.19/GPU/hr (on-demand rate $3.99/hr unchanged) Added two new serverless inference models to the model catalog: Kimi K3 ($3.00/$15.00 per 1M input/output tokens, $0.30 cached input) and Inkling Small ($0.50/$1.20 per 1M input/output tokens, $0.10 cached input)

LFM2.5-8B-A1B model removed from the Serverless Inference pricing table (previously $0.03/1M input tokens, $0.12/1M output tokens).

Added 3 new models to the Serverless Inference Chat pricing table: PrismML Ternary Bonsai 27B (free, $0.00), Inkling ($1.00 input/$4.05 output per 1M tokens), and LFM2.5-8B-A1B ($0.03 input/$0.12 output per 1M tokens)

Added Provisioned Throughput (PTU) pricing ($0.05/PTU/min, MiniMax M3 & GLM-5.2); Dedicated Inference on-demand GPU rates cut (H100 $6.49->$5.49/hr, HGX B200 $11.95->$8.99/hr)

GLM-5 model removed from serverless inference chat pricing table ($1.00 input / $3.20 output per 1M tokens). GLM-5.1 and GLM-5.2 remain available.

Together AI removed Llama 4 Maverick ($0.27/$0.85) and Kimi K2.5 ($0.50/$2.80) from the serverless inference top-tier model lineup (GLM-5 also removed). Together AI added DeepSeek V4 Pro ($1.74/$3.48 per 1M tokens) and Gemma-4-31B-it-Pearl ($0.28/$0.86) as new flagship models on serverless inference.

GPU Clusters on-demand rates cut across all hardware: H100 $5.49→$4.79, H200 $6.79→$5.99, B200 $9.95→$8.19/hour. Reserved rates also reduced 12–25% across H100, H200, and B200 tiers. DeepSeek V4 Pro serverless inference prices reduced: input from $2.10 to $1.74/1M tokens (~17% cut) and output from $4.40 to $3.48/1M tokens (~21% cut). Cached input unchanged at $0.20.

Batch API cached input prices now publicly shown for GLM-5.1 ($0.26/1M tokens) and Qwen3.5-397B-A17B ($0.35/1M tokens); both previously displayed without cached pricing.

Multiple serverless inference models increased: Qwen3.5 9B ($0.10→$0.17 in, $0.15→$0.25 out), Llama 3.3 70B ($0.88→$1.04), Llama 3 8B Instruct Lite ($0.10→$0.14) per 1M tokens. NVIDIA Nemotron 3 Ultra added to serverless inference chat models at $0.60/1M input ($0.20 cached) and $3.60/1M output tokens.

Qwen3.7-Max serverless inference pricing decreased: input from $2.50 to $1.25/1M tokens (50% cut), output from $7.50 to $3.75/1M tokens (50% cut); new cached input price of $0.13/1M tokens added.

NVIDIA HGX H200 GPU pricing made public: on-demand now $6.79/hour (was Contact us); reserved pricing now $5.95/$4.99/$4.55/hour for 7-30/31-90/91-180 days. Qwen3.7-Max model added to Serverless Inference Chat pricing at $2.50 input / $7.50 output per 1M tokens.

Gemma 3n E4B Instruct serverless inference price increased 3x from $0.02/$0.04 to $0.06/$0.12 per 1M input/output tokens.

11 new models added to serverless inference pricing: Gemma 4 31B, MiniMax M2.7, LFM2 24B A2B, Qwen3.5-397B-A17B, Qwen3-Coder-Next, Cogito v2.1 671B, Qwen3-Coder 480B A35B Instruct, Rnj-1 Instruct, gpt-oss-20B, Mixtral 8x7B Instruct v0.1. GLM-5 renamed to GLM-5.1 with price increase: input raised from $1.00 to $1.40/1M tokens (+40%), output raised from $3.20 to $4.40/1M tokens (+37.5%).

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).

MiniMax M2.5 batch API now shows $0.06 (cached) pricing annotation 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

Reserved GPU duration requirement changed from 'between 1-180 days' to 'above 6 days' 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. Removed Qwen2.5 model family from Serverless Inference catalog: Qwen3 235B A22B FP8 Throughput ($0.20/$0.60), Qwen2.5 72B ($1.20/$1.20), Qwen2.5 Coder 32B Instruct ($0.80/$0.80), Qwen2.5-VL 72B Instruct ($1.95/$8), and Qwen QwQ-32B ($1.20/$1.20).

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.