TTFT (RightRank Probe)

Speed

Time to the first token, reasoning included, of 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.

Lower is better

Rank	Model	Provider	Time
1	Cohere: Command A+	Cohere	180 ms p90 193 ms · 5 runs
2	Cohere: Command R7B (12-2024)	Cohere	183 ms p90 499 ms · 5 runs
3	Cohere: North Mini Code (free)	Cohere	195 ms p90 1.1 s · 5 runs
4	Cohere: Command R (08-2024)	Cohere	211 ms p90 252 ms · 5 runs
5	Google: Gemini 2.5 Flash LiteRetiring	Google	263 ms p90 320 ms · 5 runs
6	Mistral: Ministral 3 8B 2512	Mistral AI	279 ms p90 325 ms · 5 runs
7	Mistral: Voxtral Small 24B 2507	Mistral AI	292 ms p90 397 ms · 5 runs
8	Mistral: Ministral 3 14B 2512	Mistral AI	314 ms p90 342 ms · 5 runs
9	Mistral: Codestral 2508	Mistral AI	320 ms p90 473 ms · 5 runs
10	Mistral: Ministral 3 3B 2512	Mistral AI	343 ms p90 394 ms · 5 runs
11	Amazon: Nova Lite 1.0	Amazon	351 ms p90 490 ms · 5 runs
12	Amazon: Nova Micro 1.0	Amazon	364 ms p90 542 ms · 5 runs
13	Qwen: Qwen3 30B A3B Instruct 2507	Alibaba	369 ms p90 410 ms · 5 runs
14	Google: Gemini 3.1 Flash Lite Preview	Google	384 ms p90 555 ms · 5 runs
15	Google: Gemini 3.1 Flash Lite	Google	386 ms p90 591 ms · 5 runs
16	Qwen: Qwen3 VL 8B InstructHistorical	Alibaba	420 ms p90 456 ms · 5 runs
17	OpenAI: GPT-3.5 Turbo	OpenAI	422 ms p90 526 ms · 5 runs
18	Mistral: Mistral Large 3 2512	Mistral AI	430 ms p90 437 ms · 5 runs
19	Qwen: Qwen3 14B	Alibaba	448 ms p90 632 ms · 5 runs
20	Qwen: Qwen3.6 Flash	Alibaba	450 ms p90 470 ms · 5 runs
21	Qwen: Qwen3 30B A3BHistorical	Alibaba	457 ms p90 503 ms · 5 runs
22	Qwen: Qwen3 Next 80B A3B InstructHistorical	Alibaba	474 ms p90 601 ms · 5 runs
23	Qwen: Qwen3.5-35B-A3B	Alibaba	477 ms p90 4.1 s · 5 runs
24	Qwen: Qwen3 VL 30B A3B Instruct	Alibaba	481 ms p90 511 ms · 5 runs
25	OpenAI: GPT-5.4 Nano	OpenAI	481 ms p90 537 ms · 5 runs
26	Qwen: Qwen-Plus	Alibaba	483 ms p90 532 ms · 5 runs
27	OpenAI: GPT-4.1 Mini	OpenAI	486 ms p90 540 ms · 5 runs
28	Mistral: Mistral Medium 3	Mistral AI	488 ms p90 580 ms · 5 runs
29	Qwen: Qwen3 235B A22B Instruct 2507	Alibaba	509 ms p90 577 ms · 5 runs
30	OpenAI: GPT-4o-mini (2024-07-18)	OpenAI	511 ms p90 564 ms · 5 runs
31	Qwen: Qwen3 VL 32B InstructHistorical	Alibaba	512 ms p90 568 ms · 5 runs
32	Qwen: Qwen3 VL 235B A22B Instruct	Alibaba	512 ms p90 680 ms · 5 runs
33	Mistral: Mistral Medium 3.1	Mistral AI	518 ms p90 576 ms · 5 runs
34	Mistral: Devstral 2 2512	Mistral AI	521 ms p90 683 ms · 5 runs
35	OpenAI: GPT-4o-mini	OpenAI	533 ms p90 594 ms · 5 runs
36	Qwen: Qwen3 Next 80B A3B ThinkingHistorical	Alibaba	536 ms p90 688 ms · 5 runs
37	Mistral: Mistral Medium 3.5	Mistral AI	548 ms p90 778 ms · 5 runs
38	MiniMax: MiniMax-01	MiniMax	559 ms p90 1.1 s · 5 runs
39	Inception: Mercury 2	Inception	560 ms p90 774 ms · 5 runs
40	Google: Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)	Google	569 ms p90 711 ms · 5 runs
41	Cohere: Command R+ (08-2024)	Cohere	571 ms p90 1.6 s · 5 runs
42	MiniMax: MiniMax M3	MiniMax	572 ms p90 731 ms · 5 runs
43	Qwen: Qwen Plus 0728Historical	Alibaba	574 ms p90 588 ms · 5 runs
44	Qwen: Qwen3.5-27B	Alibaba	580 ms p90 3.4 s · 5 runs
45	OpenAI: GPT-4.1 Nano	OpenAI	596 ms p90 2.0 s · 5 runs
46	Mistral Large 2407	Mistral AI	631 ms p90 985 ms · 5 runs
47	SpaceXAI: Grok Build 0.1	SpaceXAI	634 ms p90 5.5 s · 5 runs
48	OpenAI: GPT-5.4 Mini	OpenAI	639 ms p90 675 ms · 5 runs
49	Qwen: Qwen3.7 Flash	Alibaba	647 ms p90 773 ms · 5 runs
50	Qwen: Qwen3.5-Flash	Alibaba	654 ms p90 719 ms · 5 runs

Access this data

JSON GET /api/v1

[
  {
    "rank": 1,
    "model": "Cohere: Command A+",
    "provider": "Cohere",
    "score": "179.8"
  },
  {
    "rank": 2,
    "model": "Cohere: Command R7B (12-2024)",
    "provider": "Cohere",
    "score": "183.2"
  },
  {
    "rank": 3,
    "model": "Cohere: North Mini Code (free)",
    "provider": "Cohere",
    "score": "194.7"
  }
]

Curl

$ curl -s https://rightrank.ai/api/v1/rankings/ttft-rightrank-probe.json | jq '.[:3]'

CLI brew install rightrank

$ rightrank top ttft-rightrank-probe --limit 5
$ rightrank compare claude-opus-4.6 gpt-5.4-pro --category ttft-rightrank-probe
$ rightrank value ttft-rightrank-probe  # best performance per dollar

Claude Code Skill Skill

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