Generation Tokens/sec incl. Reasoning (RightRank Probe)

Speed

All generated tokens, reasoning included, per second of the whole request, for a prompt that asks for about 500 tokens: median of 5 streamed runs through OpenRouter, pinned to the model's first-party provider, from the us-east-1 region.

Higher is better

Rank	Model	Provider	Throughput
1	Inception: Mercury 2.5	Inception	418.7 tok/s p90 435.4 tok/s · 5 runs
2	Google: Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)	Google	365.5 tok/s p90 392.6 tok/s · 5 runs
3	Inception: Mercury 2	Inception	320.4 tok/s p90 340.2 tok/s · 5 runs
4	SpaceXAI: Grok 4.20 Multi-Agent	SpaceXAI	266.8 tok/s p90 402.7 tok/s · 5 runs
5	Amazon: Nova Micro 1.0	Amazon	250.7 tok/s p90 253.4 tok/s · 5 runs
6	Google: Gemini 2.5 Flash LiteRetiring	Google	242.2 tok/s p90 255.7 tok/s · 5 runs
7	Mistral: Ministral 3 3B 2512	Mistral AI	241.5 tok/s p90 256.5 tok/s · 5 runs
8	Google: Gemini 3.1 Flash Lite Preview	Google	210.6 tok/s p90 223.8 tok/s · 5 runs
9	Google: Gemini 3.1 Flash Lite	Google	203.3 tok/s p90 231.2 tok/s · 5 runs
10	Qwen: Qwen3 Next 80B A3B ThinkingHistorical	Alibaba	190.0 tok/s p90 192.2 tok/s · 5 runs
11	Google: Gemini 3.5 Flash Lite	Google	182.6 tok/s p90 191.9 tok/s · 5 runs
12	OpenAI: o3 Mini	OpenAI	182.2 tok/s p90 187.1 tok/s · 5 runs
13	Google: Gemini 3.8 Flash	Google	177.0 tok/s p90 189.6 tok/s · 5 runs
14	Amazon: Nova Lite 1.0	Amazon	173.5 tok/s p90 188.6 tok/s · 5 runs
15	Amazon: Nova 2 Lite	Amazon	170.1 tok/s p90 171.8 tok/s · 5 runs
16	Google: Gemini 3.7 Flash	Google	168.2 tok/s p90 201.0 tok/s · 5 runs
17	OpenAI: GPT-6 Luna Pro	OpenAI	167.3 tok/s p90 187.4 tok/s · 5 runs
18	OpenAI: o4 Mini	OpenAI	167.1 tok/s p90 181.3 tok/s · 5 runs
34	↳OpenAI: o4 Mini High	OpenAI	134.9 tok/s p90 180.6 tok/s · 5 runs
19	Qwen: Qwen3.5-122B-A10B	Alibaba	163.7 tok/s p90 172.4 tok/s · 5 runs
20	Qwen: Qwen3.5-Flash	Alibaba	163.0 tok/s p90 168.4 tok/s · 5 runs
21	OpenAI: GPT-5.4 Nano	OpenAI	161.1 tok/s p90 174.9 tok/s · 5 runs
22	Qwen: Qwen3.5-35B-A3B	Alibaba	159.8 tok/s p90 171.3 tok/s · 5 runs
23	Google: Nano Banana 2 (Gemini 3.1 Flash Image Preview)	Google	157.3 tok/s p90 162.7 tok/s · 5 runs
24	OpenAI: GPT-4.1 Mini	OpenAI	157.0 tok/s p90 157.7 tok/s · 5 runs
25	Google: Nano Banana 2 (Gemini 3.1 Flash Image)	Google	156.0 tok/s p90 162.5 tok/s · 5 runs
26	OpenAI: GPT-4.1 Nano	OpenAI	155.8 tok/s p90 167.0 tok/s · 5 runs
27	Google: Gemini 2.5 FlashRetiring	Google	155.6 tok/s p90 169.7 tok/s · 5 runs
28	Perplexity: Sonar	Perplexity	152.9 tok/s p90 179.3 tok/s · 5 runs
29	Mistral: Voxtral Small 24B 2507	Mistral AI	147.4 tok/s p90 156.2 tok/s · 5 runs
30	OpenAI: GPT-5.6 Luna Pro	OpenAI	144.3 tok/s p90 162.1 tok/s · 5 runs
31	Google: Gemini 3.6 Flash	Google	140.5 tok/s p90 143.8 tok/s · 5 runs
32	Google: Gemini 3 Flash Preview	Google	137.2 tok/s p90 139.7 tok/s · 5 runs
33	OpenAI: GPT-4o-mini	OpenAI	136.0 tok/s p90 141.1 tok/s · 5 runs
35	OpenAI: GPT-6 Luna	OpenAI	133.9 tok/s p90 139.6 tok/s · 5 runs
36	Qwen: Qwen3 Coder Next	Alibaba	131.2 tok/s p90 136.9 tok/s · 5 runs
37	OpenAI: GPT-4o-mini (2024-07-18)	OpenAI	128.4 tok/s p90 139.6 tok/s · 5 runs
38	Amazon: Nova Pro 1.0	Amazon	126.1 tok/s p90 136.3 tok/s · 5 runs
39	Mistral: Mistral Medium 3	Mistral AI	125.9 tok/s p90 129.7 tok/s · 5 runs
40	Mistral: Mistral Medium 3.1	Mistral AI	124.8 tok/s p90 126.5 tok/s · 5 runs
41	Mistral: Codestral 2508	Mistral AI	124.7 tok/s p90 127.5 tok/s · 5 runs
42	Qwen: Qwen3.6 Flash	Alibaba	119.8 tok/s p90 122.4 tok/s · 5 runs
43	Mistral: Devstral 2 2512	Mistral AI	119.7 tok/s p90 127.9 tok/s · 5 runs
44	OpenAI: GPT-3.5 Turbo 16k	OpenAI	117.6 tok/s p90 123.0 tok/s · 5 runs
45	OpenAI: GPT-5.4 Mini	OpenAI	117.4 tok/s p90 120.5 tok/s · 5 runs
46	SpaceXAI: Grok Build 0.1	SpaceXAI	116.6 tok/s p90 118.4 tok/s · 5 runs
47	Qwen: Qwen3 VL 30B A3B Instruct	Alibaba	112.2 tok/s p90 114.1 tok/s · 5 runs
48	OpenAI: GPT-3.5 Turbo	OpenAI	111.5 tok/s p90 115.6 tok/s · 5 runs
49	OpenAI: GPT-3.5 Turbo Instruct	OpenAI	109.5 tok/s p90 116.6 tok/s · 5 runs
50	Qwen: Qwen3.5 397B A17B	Alibaba	108.2 tok/s p90 111.8 tok/s · 5 runs

Access this data

JSON GET /api/v1

[
  {
    "rank": 1,
    "model": "Inception: Mercury 2.5",
    "provider": "Inception",
    "score": "418.7"
  },
  {
    "rank": 2,
    "model": "Google: Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)",
    "provider": "Google",
    "score": "365.5"
  },
  {
    "rank": 3,
    "model": "Inception: Mercury 2",
    "provider": "Inception",
    "score": "320.4"
  }
]

Curl

$ curl -s https://rightrank.ai/api/v1/rankings/generation-tokens-sec-incl-reasoning-rightrank-probe.json | jq '.[:3]'

CLI brew install rightrank

$ rightrank top generation-tokens-sec-incl-reasoning-rightrank-probe --limit 5
$ rightrank compare claude-opus-4.6 gpt-5.4-pro --category generation-tokens-sec-incl-reasoning-rightrank-probe
$ rightrank value generation-tokens-sec-incl-reasoning-rightrank-probe  # best performance per dollar

Claude Code Skill Skill

# In Claude Code, use the RightRank skill:
/rightrank best model for generation-tokens-sec-incl-reasoning-rightrank-probe
/rightrank compare top 5 by value
/rightrank cheapest model scoring above 80 on generation-tokens-sec-incl-reasoning-rightrank-probe