At the median major provider in July 2026, 0.24% of requests failed with a provider 5xx and 0.46% were rate limited with a 429.

Key findings

1. 01DeepSeek had the lowest provider error rate among major providers at 0.04%.
2. 024 of 16 major providers rate limited more than 1% of requests with a 429.
3. 03The median major provider rejected 1.03% of requests before the provider call. Those errors come from the request, not the provider.
4. 04Novita changed the most since June 2026: provider errors went down by 9.4 pp.

Provider error rate by month

What changed since June 2026

• Novita10.9% → 1.5% (-9.4 pp)
• Fireworks0.2% → 8.1% (+7.9 pp)
• Vertex (Claude)5.4% → 0.8% (-4.7 pp)
• DeepSeek1.8% → 0.0% (-1.8 pp)
• Vertex (Gemini)2.5% → 3.6% (+1.1 pp)

What to do about it

• Treat 429s as a capacity problem: spread load across providers or keys rather than retrying the same endpoint.
• Set a fallback policy so a provider 5xx or 429 retries on a second provider for the same model.
• Check rejected-before-provider errors in your own logs first. They usually point to request shape or credit, not the provider.

Data

16 rows published. Rows below the floors are withheld.

Questions this answers

Which LLM provider has the fewest errors?

At the median major provider in July 2026, 0.24% of requests failed with a provider 5xx and 0.46% were rate limited with a 429.

How often do LLM APIs return 5xx errors?

At the median major provider in July 2026, 0.24% of requests failed with a provider 5xx and 0.46% were rate limited with a 429.

How often do LLM providers rate limit requests (429)?

At the median major provider in July 2026, 0.24% of requests failed with a provider 5xx and 0.46% were rate limited with a 429.

Are LLM API errors caused by the provider or the client?

At the median major provider in July 2026, 0.24% of requests failed with a provider 5xx and 0.46% were rate limited with a 429.

Method and definitions

• Errors are split by where they came from. Provider 5xx and 429 are status codes the provider returned. Other provider 4xx covers 401 to 499 excluding 429 (key, credit, model or size errors at the provider). Status 400 from the provider is treated as a malformed request and left out.
• Rejected before provider counts errors the gateway raised itself, such as invalid requests or blocked policies. They are shown separately because they say nothing about the provider.
• Only providers with at least 0.5% of published traffic in the month are charted. Vertex is split into Vertex (Gemini) and Vertex (Claude).
• 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.

Provider error rate:

Provider 5xx + provider 429 + other provider 4xx, divided by all requests to the provider.

Major provider:

At least 0.5% of published gateway traffic in the month.

Per-model numbers with the same definitions are on /rankings, refreshed weekly.

Caveats

• Error rates are per attempt. When the gateway falls back to another provider, the failed attempt still counts here even if the user got a response.
• 429 rates depend on the rate limits of the keys in use, so they partly reflect how much capacity each provider grants, not only provider health.
• A provider's rate moves with its traffic mix. A new model launch or one large workload can shift a month.

Revision history and other months

• October 2026first published later
• September 2026first published later
• August 2026first published later
• July 2026 (this page)first published 2026-11-11
• June 2026first published 2026-11-11
• May 2026first published 2026-11-11

Related

• Time to first token by providerWhich LLM provider returns the first token fastest?
• Prompt cache hit rate by providerWhat share of input tokens do LLM providers serve from prompt cache?
• Operational metrics per provider, October 2026April 2026 note
• Provider error code distribution, October 2026April 2026 note
• Policy vs direct eventual success rate, January to October 2026April 2026 note

Cite and reuse

Free to reuse under CC BY 4.0. Credit "Requesty Data" and link to this page. Cite the month URL so the numbers you quote stay fixed.

Requesty (2026). Provider error rates, July 2026. Requesty Data. https://www.requesty.ai/data/provider-error-rates/2026-07

@misc{requesty_provider_error_rates_2026_07,
  author       = {{Requesty}},
  title        = {Provider error rates, July 2026},
  year         = {2026},
  howpublished = {\url{https://www.requesty.ai/data/provider-error-rates/2026-07}},
  note         = {Requesty Data}
}

period,provider,provider_error_rate_fraction,provider_5xx_rate_fraction,provider_429_rate_fraction,provider_other_4xx_rate_fraction,router_error_rate_fraction,request_share_fraction,sample_band
2026-07,DeepSeek,0.0004,0.0004,0,0,0.0364,0.093,500+
2026-07,Sference,0.0018,0.0011,0.0007,0,0.0039,0.017,100-499
2026-07,Azure,0.0031,0.0007,0.0024,0,0.0402,0.0701,500+
…