---
id: provider-latency/2026-08
series: provider-latency
title: Time to first token by provider, August 2026
period: 2026-08
period_start: 2026-08-01
period_end: 2026-08-31
updated: 2026-11-11
license: CC BY 4.0
canonical: https://www.requesty.ai/data/provider-latency/2026-08
latest: https://www.requesty.ai/data/provider-latency
---

# Time to first token by provider, August 2026

> Sference had the fastest median time to first token among major providers in August 2026 at 55 ms. The median major provider took 1.33 s.

Question: Which LLM provider returns the first token fastest?

## Key findings

1. Fastest median TTFT: Sference at 55 ms, then Alibaba at 200 ms.
2. Highest median output speed: Vertex (Gemini) at 192 tokens per second.
3. Fastest median full streamed response: Alibaba at 1.07 s.
4. Novita moved the most since July 2026: median TTFT went from 3.71 s to 1.40 s (-2.31 s).

## Data

| Provider | Median TTFT | p90 TTFT | Median total | Median tokens/s | Streamed share | Organizations (band) |
| --- | --- | --- | --- | --- | --- | --- |
| Sference | 55 ms | 142 ms | 1.80 s | 152.700 | 74.24% | 500+ |
| Alibaba | 200 ms | 4.54 s | 1.07 s | 110.200 | 33.76% | 100-499 |
| Nebius | 542 ms | 3.20 s | 2.22 s | 130.700 | 23.46% | 500+ |
| OpenAI | 547 ms | 4.31 s | 3.63 s | 69.900 | 13.71% | 500+ |
| Azure | 608 ms | 1.64 s | 1.98 s | 52.800 | 54.69% | 500+ |
| DeepSeek | 857 ms | 1.45 s | 9.21 s | 38.700 | 50.46% | 500+ |
| DeepInfra | 930 ms | 3.83 s | 3.19 s | 73.600 | 27.53% | 500+ |
| Fireworks | 1.19 s | 3.71 s | 4.67 s | 88.300 | 34.53% | 500+ |
| Vertex (Claude) | 1.33 s | 3.08 s | 5.80 s | 77.700 | 44.49% | 500+ |
| TensorX | 1.35 s | 5.31 s | 4.96 s | 80.200 | 28.96% | 500+ |
| Novita | 1.40 s | 2.95 s | 25.3 s | 13.100 | 42.45% | 500+ |
| xAI | 1.47 s | 3.40 s | 9.67 s | 44.600 | 9.93% | 100-499 |
| Mistral | 1.55 s | 10.4 s | 3.01 s | 117.800 | 20.15% | 500+ |
| Anthropic | 1.60 s | 3.37 s | 7.27 s | 59.600 | 24.35% | 500+ |
| Bedrock | 1.74 s | 3.30 s | 6.08 s | 80.400 | 30.02% | 500+ |
| Vertex (Gemini) | 1.94 s | 9.17 s | 3.91 s | 192.400 | 5.08% | 500+ |
| NVIDIA | 2.05 s | 14.0 s | 8.48 s | 60.700 | 66.86% | 500+ |

## Method

- Time to first token and tokens per second are measured on streamed, successful requests only, with the same formulas /rankings uses. Tokens per second counts visible output tokens after the first token, with a 250 ms floor on decode time and at least 20 output tokens.
- A provider needs at least 500 streamed successes in the month for latency to be published. Only providers with at least 0.5% of published traffic are shown.
- A row is published only when it has at least 20 requests, at least 5 distinct organizations, and no single organization above 80% of its traffic. These are the same floors /rankings uses. Rows below the floors are withheld, not estimated.
- Only shares, rates, medians and organization bands are published. Absolute request, token and spend totals stay private.

## Caveats

- Latency depends on the model mix. A provider serving mostly small models will look faster than one serving large reasoning models.
- Reasoning models think before the first visible token, which raises TTFT.

## Cite as

Requesty (2026). Time to first token by provider, August 2026. Requesty Data. https://www.requesty.ai/data/provider-latency/2026-08

```bibtex
@misc{requesty_provider_latency_2026_08,
  author       = {{Requesty}},
  title        = {Time to first token by provider, August 2026},
  year         = {2026},
  howpublished = {\url{https://www.requesty.ai/data/provider-latency/2026-08}},
  note         = {Requesty Data}
}
```

Source: Requesty production gateway. License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).
Downloads: [JSON](https://www.requesty.ai/data/provider-latency/2026-08/data.json) · [CSV](https://www.requesty.ai/data/provider-latency/2026-08/data.csv) · [Markdown](https://www.requesty.ai/data/provider-latency/2026-08/data.md)