Provider count indicates ecosystem breadth and supply-side competition. Models available on more providers are less likely to suffer downtime or rate-limit bottlenecks.
These models haven't had a confirmed pricing scrape in the last 14 days.
| # | Model | Family | Provider count | Providers | Last updated |
|---|---|---|---|---|---|
| 01 | Llama 3.1 70B Instructstale | llama | 1 providers | 1 | May 27 |
| 02 | Qwen 2.5 72B Instructstale | qwen | 1 providers | 1 | May 27 |
| 03 | DeepSeek V3stale | deepseek | 1 providers | 1 | May 27 |
| 04 | Llama 3.1 8B Instructstale | llama | 1 providers | 1 | May 27 |
| 05 | DeepSeek R1 Distill Llama 70Bstale | deepseek | 1 providers | 1 | May 27 |
Provider count is a proxy for supply diversity and pricing competition. When a model is hosted by many providers — say, Llama 3.3 70B, which is available across a dozen or more inference providers — you can switch without changing your model or prompt format, bid providers against each other on price, and avoid single-provider lock-in. If your primary provider goes down, degrades service, or raises prices, you have concrete alternatives already benchmarked. Models available from only one or two providers give you less leverage and more operational risk.
A model with a single provider is either very new (adoption hasn't had time to spread), proprietary to that provider (the weights aren't publicly available), or niche enough that infrastructure operators haven't prioritized it. All three cases carry operational risk: you're dependent on that provider's uptime, pricing, and roadmap. For new open-weights models, low provider count is temporary — well-regarded models typically reach 5+ providers within 60–90 days of weight release. For proprietary models, single-provider availability is a permanent feature, not a gap, and should be factored into your vendor dependency assessment.
Roughly, yes. Models with high provider counts have typically been in production long enough for the community to validate them and for operators to invest in serving infrastructure. They're also more likely to be supported by popular inference frameworks and quantization tooling, which lowers a provider's cost to offer them. That said, provider count doesn't guarantee any individual provider's uptime or pricing stability — it only tells you how many alternatives exist. A model on 15 providers can still see all 15 degrade simultaneously if they share a common upstream dependency or hardware supply constraint.
Yes. For provider-count purposes, `meta/llama-3.3-70b-instruct` and `meta/llama-3.3-70b-instruct-fp8` are tracked as distinct rows. A provider offering only the FP8 variant is not counted toward the provider total for the base FP16 model. This matters because some budget-oriented providers only serve quantized variants, which changes the quality-cost profile of the offering. The leaderboard shows provider counts for each canonical model ID and its tracked quantization variants separately so you can see whether the pricing and availability you're comparing reflects the same underlying weights and precision.