LLM7
llm7.io · Inference · 1 record · JSON
LLM7 Inference
caveatanonymousno cardMrequest-rate
| Free tier | no key required for the ids probed; /v1/models lists 68 ids, none with a zero price unit tokens; metered on request-count; the catalogue carries a per-model pricing object in USD per 1M tokens and every one is above zero, which makes the free access a quota rather than a price |
|---|---|
| Price | none stated |
| Per-object cap | not applicable: an inference endpoint stores nothing for the caller |
| Retention | not applicable |
| Egress | the response is the payload; no separate egress term is published |
| Identity | anonymous no Authorization header sent and none required for the calls that answered. The prober's per-endpoint track record reports 380 of 1,029 calls answered over 14 days, 37 per cent, and 380 of the 556 calls LLM7 accepted for a model it offers, 68 per cent |
| Limits |
no rate-limit header was returned on a live call, and the ceiling was not probed by provoking 429s, which would spend an allowance belonging to someone else's service. |
| Binding | request-rate |
| Interface | rest endpoint: https://api.llm7.io/v1/chat/completions |
| Tooling | any OpenAI-compatible client; curl against POST {base}/chat/completions |
| Composes with | pairs with a free-standing agent; the undocumented rate ceiling is the scarce link, because a retry loop is needed and the ceiling it loops against is not published |
| Reliability | one id was refused with 429 'Rate limit exceeded. Retry after 1 seconds' in the second run after answering in the first, eight minutes apart |
| Notes | The cleanest available ordering in the sweep: nemotron-3-nano:30b measured 53.12 and 86.43 tok/s and codestral-latest 127.60 and a 429. The two ids recorded as listed-but-refusing, GLM-5.3-Flash and DeepSeek-V4-Flash-0731, do not appear in /v1/models at all; the catalogue carries glm-5.3 and deepseek-v4-flash:0731, so a 400 on the unpublished slug proves nothing about the catalogue's honesty. |
| Evidence rung | measured four ids were called twice. The answer rate is third-party; the 68 per cent figure is the prober's own accepted-only correction and both numbers are quoted because the headline one understates the endpoint. No quota is published, so the ceiling is measured only as 'it answered now'. |
| Checked | 2026-10-23 (live); stale on 2026-11-22 |
Evidence (2)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | probe:https://api.llm7.io/v1/chat/completions | nemotron-3-nano:30b answered 200 with the model echoed and the needle present, 67 completion tokens, 672 ms in run 1 and 631 ms in run 2; codestral-latest answered in run 1 and returned 429 rate_limit_exceeded in run 2 |
| 2026-10-22 | api.llm7.io/v1/models | 68 model ids each carrying a pricing object with input, output and cached_input in USD per 1M tokens; no id is priced at zero; glm-5.3 and deepseek-v4-flash:0731 are listed, while the two slugs the sweep sent, GLM-5.3-Flash and DeepSeek-V4-Flash-0731, are absent |
Unknown or withheld values (6)
$.probe.records[0].tps_basisprocess$.probe.records[1].tps_basisprocess$.probe.records[2].tps_basisprocess$.probe.records[3].tps_basisprocess$.probe.records[4].tps_basisprocess$.retention.statednot-applicable
Corrections
- 2026-10-22 The prober's answer rates were right and the reading of them was wrong