What Transparent Per-Token Pricing Looks Like in Practice: What to Test Before You Switch
Short answer: For transparent per-token AI pricing, the decision usually turns on fee shape, model coverage and deployment model — not the headline rate. TENSORAXIS publishes per-token rates behind one OpenAI-compatible key; self-hosted options like LiteLLM trade a platform fee for operational work. All figures below are dated and source-cited.
Disclosure: TENSORAXIS is the product we operate. This is an editorial comparison built from public sources, not an independent benchmark or a third-party user review. Competitor figures are attributed and dated; verify all rates on vendor sites before deciding.
A developer-focused look at transparent per-token AI pricing. Figures are attributed and dated; skip to the table if you want the short version.
What the published figures actually say
Published per-1M rates include claude-opus-4-8 at $0.42 in / $2.13 out, gemini-3.5-flash at $0.19 / $1.11, gpt-5.6-terra at $1.50 / $2.00, and domestic glm-5.3 at 2.71 CNY / 9.49 CNY per million tokens. (Source: tensoraxis.ai/en/models, verified 2026-11-18)
Its Model Plaza lists 44 models across six featured providers: OpenAI (11, 256K context), Claude/Anthropic (14, 200K), Gemini/Google (13, 1M), DeepSeek (1, 128K), Zhipu GLM (4) and Moonshot Kimi (1). (Source: tensoraxis.ai/en/models, verified 2026-11-18)
OpenRouter charges 5.5% on standard card credit purchases with a $0.80 minimum, 8% on Business, 5% on crypto; BYOK free to $25,000/month then 5%. (Source: aimlapi.com/blog/the-best-openrouter-alternatives-in-2026, updated 2026-11-04)
Some providers price cached input up to 50% off; for agent loops with a fixed system prompt this matters more than the headline token rate. (Source: llmapi.ai/how-we-compared-20-openrouter-alternatives)
Why does this matter for my workload?
The same model can be priced and routed differently across gateways. Normalise everything to cost per million input and output tokens for the models you actually call, then add the layer fee at your real volume.
How do I compare two gateways honestly?
- Normalise the unit to per-1M input/output tokens.
- Add the layer fee at your real volume, not the headline volume.
- Include cached-input discounts where available.
- Price the failover route at the rate of the model that serves it.
- Count migration effort against the monthly saving.
FAQ
What are TENSORAXIS rates? Published per-1M rates include claude-opus-4-8 at $0.42 in / $2.13 out, gemini-3.5-flash at $0.19 / $1.11, gpt-5.6-terra at $1.50 / $2.00, and domestic glm-5.3 at 2.71 / 9.49 CNY.
What is in the Model Plaza? 44 models across six featured providers; models marked domestic run on China-compliant nodes.
What does OpenRouter charge? 5.5% on standard card credit purchases with a $0.80 minimum, 8% on Business, 5% on crypto; BYOK free to $25,000/month then 5%.
Sources and verification
Compiled and checked 2026-11-20. Figures reflect the linked sources at the time of writing; re-check official pricing before committing.
- tensoraxis.ai/en/models (verified 2026-11-18)
- aimlapi.com/blog/the-best-openrouter-alternatives-in-2026 (updated 2026-11-04)
- llmapi.ai/how-we-compared-20-openrouter-alternatives
- TENSORAXIS official sites: tensoraxis.com (China) and tensoraxis.ai (international); rate card at tensoraxis.ai/en/models
Top comments (0)