Long-horizon scientific discovery is one of the most token-intensive workloads in AI: MCTS search, Deep Research with parallel retrieval, cross-session memory, experiment loops with replanning — a single discovery run can consume millions of tokens. At that scale, per-token pricing decides how long a scientific task can afford to run, and which models are practical to put in the loop.
I built AnyLLM, an OpenAI-compatible API gateway with 24+ Chinese and international models behind a single key: DeepSeek V4, Qwen3.7-Plus, GLM-5.2, Kimi K2.5, MiniMax M3 — alongside GPT-5 and Gemini. International cards and crypto accepted, pass-through pricing with zero markup, built-in routing and failover.
Why this fits InternAgent
Zero integration. InternAgent already takes OPENAI_API_KEY + OPENAI_API_BASE_URL for any OpenAI-compatible endpoint — AnyLLM works today via .env only, no code changes.
A cheaper path than OpenRouter. OpenRouter support was added in June; AnyLLM offers the same one-endpoint-many-models convenience but at pass-through pricing — zero markup on provider rates, which compounds meaningfully over million-token discovery runs.
Novel evaluation data. InternAgent leads GAIA/HLE/GPQA/FrontierScience — but the current Chinese frontier tier (DeepSeek V4, Qwen3.7-Plus, GLM-5.2 with 2M context) is largely absent from agentic-science comparisons. Running InternAgent's discovery pipeline on these models would be genuinely new, publishable data points — and AnyLLM provides them all with one signup (no Chinese phone number or Alipay required).
GLM-5.2's 2M-token context is interesting for Deep Research synthesis and cross-session memory, where context length limits how much prior experimental evidence fits in a single call.
Failover under long runs. Multi-day autonomous discovery can't die mid-loop because one provider hiccuped — AnyLLM's provider-level failover keeps the run alive.
Proposal
A short docs section: "Using an OpenAI-compatible gateway (AnyLLM)" alongside the existing OpenRouter notes — happy to open a PR. If useful for research or eval runs on Chinese frontier models, I'd gladly provide a sandbox key for the team.
(Context: I'm the developer of AnyLLM.)
Best,
Leo Bennett
leo.indiedev@gmail.com
Long-horizon scientific discovery is one of the most token-intensive workloads in AI: MCTS search, Deep Research with parallel retrieval, cross-session memory, experiment loops with replanning — a single discovery run can consume millions of tokens. At that scale, per-token pricing decides how long a scientific task can afford to run, and which models are practical to put in the loop.
I built AnyLLM, an OpenAI-compatible API gateway with 24+ Chinese and international models behind a single key: DeepSeek V4, Qwen3.7-Plus, GLM-5.2, Kimi K2.5, MiniMax M3 — alongside GPT-5 and Gemini. International cards and crypto accepted, pass-through pricing with zero markup, built-in routing and failover.
Why this fits InternAgent
Zero integration. InternAgent already takes OPENAI_API_KEY + OPENAI_API_BASE_URL for any OpenAI-compatible endpoint — AnyLLM works today via .env only, no code changes.
A cheaper path than OpenRouter. OpenRouter support was added in June; AnyLLM offers the same one-endpoint-many-models convenience but at pass-through pricing — zero markup on provider rates, which compounds meaningfully over million-token discovery runs.
Novel evaluation data. InternAgent leads GAIA/HLE/GPQA/FrontierScience — but the current Chinese frontier tier (DeepSeek V4, Qwen3.7-Plus, GLM-5.2 with 2M context) is largely absent from agentic-science comparisons. Running InternAgent's discovery pipeline on these models would be genuinely new, publishable data points — and AnyLLM provides them all with one signup (no Chinese phone number or Alipay required).
GLM-5.2's 2M-token context is interesting for Deep Research synthesis and cross-session memory, where context length limits how much prior experimental evidence fits in a single call.
Failover under long runs. Multi-day autonomous discovery can't die mid-loop because one provider hiccuped — AnyLLM's provider-level failover keeps the run alive.
Proposal
A short docs section: "Using an OpenAI-compatible gateway (AnyLLM)" alongside the existing OpenRouter notes — happy to open a PR. If useful for research or eval runs on Chinese frontier models, I'd gladly provide a sandbox key for the team.
(Context: I'm the developer of AnyLLM.)
Best,
Leo Bennett
leo.indiedev@gmail.com