Muse Spark 1.3/Open 816B
Long-context multi-turn realistic agentic scenario (AgentX)
Hyperscaler cost ↓ Lower is better
Source: InferenceX & SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model
| Model · scenario | B200 | MI355X | B300 | GB200 NVL72 | GB300 NVL72 |
|---|---|---|---|---|---|
Muse Spark 1.3/Open 816BLong-context multi-turn realistic agentic scenario (AgentX) | SGLang · FP4 | SGLang · FP4 | SGLang · FP4 | —no exact @100 result | Dynamo SGLang · FP4 |
DeepSeek V4.1 Flash 552BLong-context multi-turn realistic agentic scenario (AgentX) | —no exact @100 result | —no exact @100 result | |||
Kimi K3.1 Pro 2.8TLong-context multi-turn realistic agentic scenario (AgentX) | —cannot reach @100 | vLLM · FP4 | SGLang · FP4 | Dynamo SGLang · FP4 | Dynamo vLLM · FP4 |
MiniMax M3.1 Flash 428BLong-context multi-turn realistic agentic scenario (AgentX) | SGLang · FP4 | SGLang · FP4 | SGLang · FP4 | Dynamo vLLM · FP4 | SGLang · FP4 |
GLM5.2/GLM5.3 744BLong-context multi-turn realistic agentic scenario (AgentX) | —no exact @100 result | Dynamo SGLang · FP4 | |||
Qwen3.5 397BLong-context multi-turn realistic agentic scenario (AgentX) | SGLang · FP4 | Dynamo SGLang · FP4 | Dynamo SGLang · FP4 | ||
Qwen3.8 Flash Next 176BLong-context multi-turn realistic agentic scenario (AgentX) | —no exact @100 result | SGLang · FP4 | SGLang · FP4 | —no data for this scenario | |
Qwen3.5 397B8K/1K | SGLang · FP4 | SGLang · FP4 | Dynamo SGLang · FP8 |
Long-context multi-turn realistic agentic scenario (AgentX)
Long-context multi-turn realistic agentic scenario (AgentX)
Long-context multi-turn realistic agentic scenario (AgentX)
Long-context multi-turn realistic agentic scenario (AgentX)
Long-context multi-turn realistic agentic scenario (AgentX)
Long-context multi-turn realistic agentic scenario (AgentX)
Long-context multi-turn realistic agentic scenario (AgentX)
8K/1K
Current cost and change versus the latest validated platform result 30–60 days earlier.
Platforms without a valid 30-day comparison show current cost only.
If a chip does not have FP4 spec decoding available, the next best available configuration is used.