ChatAnywhere GPT_API_free (free GitHub relay)

The free tier auto-refreshes 50000 free credits per week (previously recorded as 10000 per day), plus a cap of 100 requests per day; there's also an unchanged hard rate limit of 200 requests/day per IP & key (chat and embedding models counted separately)

Re-checked 2026-09-18 (GitHub README original text): 'the free tier gets a 50000-credit free quota per week, plus a 100-request-per-day cap' — the quota cycle has changed from the previously recorded 'daily 10000 credits, tiered by model at 5/30/100 uses' to a new 'weekly 50000 credits + 100 requests/day' framework; the per-model tiered use-counts aren't re-listed at the top of the page this round. The 200-requests/day/IP&key rate limit (chat and embedding models counted separately) is unchanged; the relay endpoints api.chatanywhere.tech (direct access from mainland China, no proxy needed) and api.chatanywhere.org (overseas) are unchanged; the vendor still emphasizes that the free key is limited to personal non-commercial use, with commercial use strictly prohibited.

⚠️An open-source GitHub project maintained for years (41k+ stars, 152+ commits); the free quota is subsidized by paid-tier revenue, with the vendor explicitly stating that 'offering free keys as a pure public good isn't a sustainable model — paid keys are needed to cover day-to-day costs.' DeepSeek and other models are sourced through official channels like Volcengine, while the GPT series' exact upstream source isn't fully disclosed; a third-party risk assessment rates it medium risk (relay architecture, unverifiable upstream account-sourcing compliance). This is a third-party relay: your data passes through its servers, so don't send sensitive content; there's no guarantee of stability or continued operation.