The Known Good Updated 9 Sep 2026 Subscribe
Providers· Inference host· 45 tracked models

Parasail — pricing & performance

Compare models All providers Data sources
Models served
45
of 1,111 tracked models
Mean output speed
43
tokens per second, averaged over
37 measured endpoints
Mean latency
1.0s
time to first token, averaged over
37 measured endpoints
Cheapest blended
$0.030
per 1M tokens at a 3:1
input:output blend
Output speed on Parasail

Output tokens per second for the 37 of 45 models on Parasail with published throughput telemetry, the most recent figure OpenRouter publishes for each host. The full table below lists all 45, measured or not. Higher is better. We do not run these measurements ourselves.

⇩
94
91
86
70
62
61
59
59
57
57
55
55
The Known Good
45 models on Parasail
Full leaderboard →
Model Creator Input $/M Output $/M Blended Speed TTFT Context
Arcee AI: Trinity Large Thinking arcee-ai $0.22 $0.85 $0.38 — — 262k
ByteDance: UI-TARS 7B ByteDance $0.10 $0.20 $0.12 7 1.6s 128k
DeepSeek: DeepSeek V4 Flash 0423 DeepSeek $0.14 $0.28 $0.17 46 0.9s 1.05M
DeepSeek: DeepSeek V4 Flash 0731 DeepSeek $0.14 $0.28 $0.17 23 1.1s 1.05M
DeepSeek: DeepSeek V4 Pro 0423 DeepSeek $1.74 $3.48 $2.17 52 0.9s 1.05M
DeepSeek: DeepSeek V4 Pro 0813 DeepSeek $1.32 $3.96 $1.98 55 1.2s 1.05M
Google: Gemma 3 27B Google $0.080 $0.45 $0.17 26 1.1s 131k
Google: Gemma 4 26B A4B Google $0.13 $0.40 $0.20 49 0.6s 262k
Google: Gemma 4 31B Google $0.15 $0.40 $0.21 21 1.3s 262k
Meta: Llama 3.2 3B Instruct Meta $0.050 $0.33 $0.12 32 0.6s 131k
Meta: Llama 3.3 70B Instruct Meta $0.22 $0.50 $0.29 31 0.6s 131k
Meta: Llama 4 Maverick Meta $0.35 $1.00 $0.51 37 0.6s 524k
Meta: Muse Glimmer 30B Meta $0.30 $1.10 $0.50 — — 131k
MiniMax: MiniMax M2.5 MiniMax $0.30 $1.20 $0.53 — — 197k
MiniMax: MiniMax M3 MiniMax $0.30 $1.20 $0.53 55 0.8s 1.05M
Mistral: Mistral Nemo Mistral $0.030 $0.030 $0.030 40 0.5s 131k
Mistral: Mistral Small 3.2 24B Mistral $0.090 $0.30 $0.14 22 0.5s 131k
MoonshotAI: Kimi K2.6 Moonshot AI $0.75 $3.50 $1.44 70 1.3s 262k
MoonshotAI: Kimi K2.7 Code Moonshot AI $0.76 $3.50 $1.45 — — 262k
MoonshotAI: Kimi K3 Moonshot AI $3.00 $15.00 $6.00 62 1.2s 1.05M
MythoMax 13B gryphe $0.080 $0.11 $0.087 91 0.3s 4k
OpenAI: gpt-oss-120b OpenAI $0.10 $0.75 $0.26 86 0.5s 131k
OpenAI: gpt-oss-20b OpenAI $0.030 $0.15 $0.060 61 0.5s 131k
Qwen: Qwen2.5 VL 72B Instruct Alibaba $0.80 $1.00 $0.85 3 0.6s 128k
Qwen: Qwen3 235B A22B Instruct 2507 Alibaba $0.14 $0.80 $0.30 21 0.6s 131k
Qwen: Qwen3 Coder Next Alibaba $0.12 $0.80 $0.29 52 1.1s 262k
Qwen: Qwen3 Next 80B A3B Instruct Alibaba $0.10 $1.10 $0.35 40 0.9s 262k
Qwen: Qwen3 VL 235B A22B Instruct Alibaba $0.21 $1.90 $0.63 9 1.4s 131k
Qwen: Qwen3 VL 8B Instruct Alibaba $0.25 $0.75 $0.38 20 1.9s 262k
Qwen: Qwen3.5 397B A17B Alibaba $0.50 $3.60 $1.27 45 0.9s 262k
Qwen: Qwen3.5-35B-A3B Alibaba $0.15 $1.00 $0.36 57 0.5s 262k
Qwen: Qwen3.5-9B Alibaba $0.10 $0.25 $0.14 27 4.4s 262k
Qwen: Qwen3.6 35B A3B Alibaba $0.15 $1.00 $0.36 40 0.6s 262k
Qwen: Qwen3.8 27B Alibaba $0.24 $2.20 $0.73 45 1.4s 262k
Sao10K: Llama 3 8B Lunaris sao10k $0.040 $0.050 $0.043 46 0.5s 8k
TheDrummer: Cydonia 24B V4.1 thedrummer $0.30 $0.50 $0.35 15 1.2s 131k
TheDrummer: Rocinante 12B thedrummer $0.25 $0.50 $0.31 — — 66k
TheDrummer: Skyfall 36B V2 thedrummer $0.55 $0.80 $0.61 59 0.6s 33k
TheDrummer: UnslopNemo 12B thedrummer $0.40 $0.40 $0.40 42 0.8s 1.02M
Xiaomi: MiMo-V2.5 Xiaomi $0.14 $0.28 $0.17 — — 1.05M
Z.ai: GLM 5 Z.ai $1.00 $3.20 $1.55 — — 203k
Z.ai: GLM 5.1 Z.ai $1.40 $4.40 $2.15 — — 203k
Z.ai: GLM 5.2 Z.ai $1.40 $4.40 $2.15 94 0.9s 262k
Z.ai: GLM 5.3 Z.ai $1.40 $4.40 $2.15 59 0.8s 1.05M
Z.ai: GLM 5.3 Flash Z.ai $0.15 $0.50 $0.24 57 1.3s 1.05M
The Known Good
Source. Prices, endpoint availability and throughput for Parasail are ingested from OpenRouter; model metadata is enriched from models.dev. A blank cell means the figure was not published for that endpoint — it never means zero. Blended price is (3 × input + output) ÷ 4, the same 3:1 blend used everywhere on this site; see methodology. Quality scores are not provider-specific: a model scores the same wherever it is hosted, so quality lives on the model pages.