TTFT

Speed

Time to First Token — latency before generation begins

Lower is better

Rank	Model	Provider	Time
1	Tiny Aya Global	Cohere	206 ms
2	Granite 4.2 3B	IBM	234 ms
3	Cohere: Command A	Cohere	248 ms
4	North Mini Code	Cohere	260 ms
5	Granite 4.2 30B	IBM	274 ms
6	NVIDIA Nemotron Nano 9B V2 (Reasoning)	NVIDIA	336 ms
7	IBM: Granite 4.2 8B	IBM	346 ms
8	microsoft/Phi-4-mini-instruct	Microsoft	346 ms
9	Nemotron 3 Nano Omni 30B A3B Reasoning	NVIDIA	360 ms
10	NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)	NVIDIA	365 ms
11	NVIDIA: Nemotron 3.5 Lightning	NVIDIA	402 ms
12	gpt-oss-20b (high)	OpenAI	406 ms
13	gpt-oss-20b (low)	OpenAI	408 ms
14	Celeris-1	Celeris	417 ms
15	NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)	NVIDIA	422 ms
16	Qwen3 Next 80B A3B (Reasoning)	Alibaba	448 ms
17	Qwen3.5 9B (Non-reasoning)	Alibaba	448 ms
18	Muse Glimmer (high)	Meta	470 ms
19	Gemma 4 E4B (Reasoning)	Google	472 ms
20	Ministral 3 3B	Mistral AI	479 ms
21	Gemma 4 E4B (Non-reasoning)	Google	497 ms
22	gpt-oss-120b (low)	OpenAI	501 ms
23	Ministral 3 8B	Mistral AI	504 ms
24	gpt-oss-120b (high)	OpenAI	508 ms
25	Qwen: Qwen3 Next 80B A3B InstructHistorical	Alibaba	573 ms
26	Quasar 438B (max, based on GLM-5.2)	Multiverse Computing	578 ms
27	Claude Sonnet 5.5 (Adaptive Reasoning, Low Effort, Default Fallback)	Anthropic	630 ms
28	Gemma 4 26B A4B (Non-reasoning)	Google	631 ms
29	Mistral: Mistral Medium 3.5	Mistral AI	634 ms
30	Mistral Small 4 (Non-reasoning)	Mistral AI	636 ms
31	Qwen3.5 4B (Reasoning)	Alibaba	638 ms
32	Meta: Llama 4 Maverick	Meta	644 ms
33	Llama 3.3 Instruct 70B	Meta	649 ms
34	Agnes 2.5 Pro Beta	Sapiens AI	649 ms
35	Nova Micro	Amazon	649 ms
36	Qwen3.5 4B (Non-reasoning)	Alibaba	666 ms
37	Gemma 4 26B A4B (Reasoning)	Google	681 ms
38	Hermes 4 - Llama-3.1 405B (Reasoning)	Nous Research	701 ms
39	Ministral 3 14B	Mistral AI	705 ms
40	Hermes 4 - Llama-3.1 405B (Non-reasoning)	Nous Research	710 ms
41	Claude Fable 5.1 (Adaptive Reasoning, Low Effort, Default Fallback)	Anthropic	715 ms
42	Mistral Small 4 (Reasoning)	Mistral AI	736 ms
43	GPT-5.5 Instant (June 2026)	OpenAI	745 ms
44	Nova 2.0 Pro Preview (Non-reasoning)	Amazon	747 ms
45	Reka Flash 3	Reka AI	755 ms
46	NVIDIA Nemotron Nano 9B V2 (Non-reasoning)	NVIDIA	761 ms
47	GPT-5.6 Terra (Non-reasoning)	OpenAI	769 ms
48	Llama 3.2 Instruct 11B (Vision)	Meta	771 ms
49	Nova 2.0 Lite (Non-reasoning)	Amazon	792 ms
50	NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)	NVIDIA	795 ms

Access this data

JSON GET /api/v1

[
  {
    "rank": 1,
    "model": "Tiny Aya Global",
    "provider": "Cohere",
    "score": "206.0"
  },
  {
    "rank": 2,
    "model": "Granite 4.2 3B",
    "provider": "IBM",
    "score": "234.0"
  },
  {
    "rank": 3,
    "model": "Cohere: Command A",
    "provider": "Cohere",
    "score": "248.0"
  }
]

Curl

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

CLI brew install rightrank

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

Claude Code Skill Skill

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