--- title: "smaapi Overseas Model Access Index: Measured Availability of Connected Models | smaapi" canonical: https://www.smaapi.com/en/reports/model-access-benchmark lang: en updated: 2026-07-26 source: https://www.smaapi.com --- # smaapi Overseas Model Access Index: Measured Access of Connected Models The smaapi Overseas Model Access Index is a measured record of the overseas LLMs x access channels the SMA gateway (www.smaapi.com) is actually connected to: it reports availability and sample count per model, obtained by active probing rather than vendor claims. Only models above the publish threshold are listed; figures are injected by the build pipeline and never hand-filled; unconnected or under-sampled models do not appear. ## 1. How are these numbers measured? - Data source: active probing and production monitoring by the SMA gateway; injected by the build pipeline, never hand-filled. - Window: current probe batch (rolling-window statistics not yet enabled), as of 2026-07-26. - P95 definition: v1 探活可用率版:不含延迟分位数;P50/P95 待网关补 per-request 延迟记录后由 v2 发布。 - Publish threshold: a model x channel with a sample count below 30 is excluded from the main table (under-sampled figures are not publishable). - How it's built: generated by the build pipeline; every number traces back to a row in the gateway export. - Machine-readable: the same data is published as a [schema.org Dataset](https://schema.org/Dataset) (defined there as "A body of structured information describing some topic(s) of interest.") and served at a [JSON endpoint](https://www.smaapi.com/data/model-access-index.json), so the definitions travel with the numbers when they are quoted elsewhere. ## 2. Which models are connected, and what did they measure? | Model | Channel | Availability | P50 (ms) | P95 (ms) | Samples | As of | |---|---|---|---|---|---|---| | sma-balanced | 云端直连(探活口径) | 100% | — | — | 48 | 2026-07-26 | | sma-domestic-code | 云端直连(探活口径) | 100% | — | — | 48 | 2026-07-26 | | sma-domestic-fast | 云端直连(探活口径) | 100% | — | — | 48 | 2026-07-26 | | sma-fast | 云端直连(探活口径) | 100% | — | — | 48 | 2026-07-26 | | sma-overseas-claude | 云端直连(探活口径) | 100% | — | — | 48 | 2026-07-26 | | sma-overseas-gpt | 云端直连(探活口径) | 100% | — | — | 48 | 2026-07-26 | | sma-overseas-pro | 云端直连(探活口径) | 100% | — | — | 48 | 2026-07-26 | | sma-strong-reasoner | 云端直连(探活口径) | 100% | — | — | 48 | 2026-07-26 | This period is the "availability-only" edition: latency percentiles (P50/P95) will ship once the gateway records per-request latency, so the latency columns are left blank (missing, not filled). Availability comes from active probing; see methodology above. The columns are left blank rather than filled with an average, because an average flattens exactly the tail that enterprise traffic runs into. As Google's SRE team puts it in the chapter on Service Level Objectives: > "Monitoring and alerting based only on the average latency would show no change in behavior over the course of the day, when there are in fact significant changes in the tail latency […]" Source: [Google SRE Book · Service Level Objectives](https://sre.google/sre-book/service-level-objectives/) (bracketed ellipsis ours) So the latency column either carries P50/P95 percentiles or stays empty — no mean value standing in for them. ## 3. Can this availability be read as an SLA? No — the two are measured differently. A cloud provider's SLA is a contractual, creditable commitment. Amazon Bedrock's SLA, for instance, defines availability like this: > "'Availability' is calculated for each 5-minute interval as the percentage of Requests processed by Amazon Bedrock that do not fail with Errors and relate solely to the Amazon Bedrock APIs for models that are available on Amazon Bedrock." "'Monthly Uptime Percentage' for a given AWS region is calculated as the average of the Availability for all 5-minute intervals in a monthly billing cycle." Source: [Amazon Bedrock Service Level Agreement](https://aws.amazon.com/bedrock/sla/) Three differences from this page: unit of measurement — an SLA aggregates real production requests over 5-minute intervals, this page counts probe rounds; scope — an SLA is computed per AWS region over a monthly billing cycle and excludes the SLA Exclusions it lists, this page is per model x channel over the current probe batch (rolling-window statistics not yet enabled); force — an SLA is a claimable term, this page is a measurement record. The figures here tell you whether a route is currently reachable and steady; they are not an availability commitment. Contractual availability terms are per enterprise contract. ## 4. Why are cost figures not published yet? The first version does not show cross-channel cost comparisons. These will ship as a separate version only once real billing data, currency, FX date, token accounting, and sample thresholds are all fixed. Left blank here rather than filled with estimates. ## 5. How are removed models recorded? Each monthly release records the models added or removed and why. The record of a model removed for failing probes is itself evidence of the index's honesty — removals are written down, not quietly dropped. Trademarks: OpenAI, Anthropic, Google, and their model names belong to their respective holders. SMA (www.smaapi.com) has no official partnership with these trademark holders; model names are used nominatively. Amazon Web Services and Amazon Bedrock are trademarks of Amazon.com, Inc. or its affiliates; this page cites their public documentation nominatively. Availability and latency are measured statistics and do not constitute an SLA; SLA terms are per enterprise contract. About the name: smaapi (the SMA gateway) is the enterprise AI gateway built by Slime Mould Tech — SMA stands for Slime Mould Architecture. smaapi is unrelated to the simple moving average indicator in finance, to the solar inverter vendor SMA Solar Technology AG, or to the SMA coaxial connector standard that share the acronym. Related: [High-availability access to overseas LLMs](https://www.smaapi.com/en/overseas-llm-high-availability) · [Compliant access to overseas LLMs](https://www.smaapi.com/en/compliant-overseas-llm-access) · [What is an enterprise LLM gateway](https://www.smaapi.com/en/what-is-llm-gateway) --- 本文为 https://www.smaapi.com/en/reports/model-access-benchmark 的 markdown 镜像,内容与该页一致,引用请指向 canonical 地址。