Together AI

Limited-time free promo for Muse Spark Open (was $0.00 in/out, valid until 10/18/2026) removed after expiry; standard $1.25 input / $4.25 output pricing remains.

New model Reflection Beam added to serverless inference pricing at $0.40 input ($0.04 cached) and $1.20 output per 1M tokens

Serverless model rate card adds Muse Spark Open ($1.25 in / $4.25 out, $0.15 cached) and NVIDIA Nemotron 3.5 Super VL ($0.15 in / $0.50 out, $0.02 cached) Muse Spark Open promo row added: input $1.00 and output $5.00 become $0.00 until 10/18/2026; site banner offers free trial through 10/18 12 p.m. PT

Kimi K3 cached input token price in the Serverless Inference (Chat) pricing table rose from $0.27 to $0.30 per 1M tokens; standard input $3.00 and output $15.00 unchanged.

Qwen3.8 Flash serverless price up from $0.09 input/$0.28 output to $0.15/$0.47 per 1M tokens. Kimi K3 PROMO pricing added: $2.70 input/$13.50 output (from $3.00/$15.00) per 1M tokens, valid until 10/11/2026.

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

NVIDIA HGX B300 GPU pricing became publicly visible in the GPU Clusters table: on-demand $9.99/GPU/hr and preemptible $4.99/GPU/hr, replacing "—" placeholders (reserved capacity remains Contact us)

Qwen3.8 Flash and Qwen3.7-Max serverless inference pricing decreased per 1M tokens: Qwen3.8 Flash $0.15/$0.47 to $0.09/$0.28; Qwen3.7-Max $2.50/$7.50 to $1.50/$4.50 (cached $0.25 to $0.30).

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)

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 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

Serverless Inference model catalog updated: new model 'Qwen3.8 Flash' added ($0.15/$0.47 per 1M tokens); 'Gemma 3n E4B Instruct' model removed (was $0.06/$0.12 per 1M tokens).

Qwen3.8-2.4T-A95B serverless inference pricing decreased: input $2.50 to $2.00, output $6.25 to $6.00, cached input $0.50 to $0.25 per 1M tokens. New GLM-5.3 model added to serverless inference pricing at $1.40 input / $4.40 output / $0.26 cached input per 1M tokens.

Serverless chat model catalog updated: added GLM-5.3-Flash ($0.15/$0.50 per 1M tokens); removed DeepSeek V4 Pro, NVIDIA Nemotron 3 Ultra, Kimi K2.7 Code, and Gemma-4-31B-it-Pearl.

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. 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)

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) 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)

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.