—

LLM Value IndexLLM VALUE INDEX · 2026

Large-model API platforms include vendor-native APIs, aggregators and hosted inference gateways — all billed per token. We just compare prices and plans and score how many tokens you get per dollar.

— — — As of —

Value ranking

VALUE INDEX

Scores are relative — the leader is 100, others scale proportionally. More tokens per dollar means a higher score. Adjust the mix in the calculator below. Cheapest isn’t always most reliable — read the notes below. Stations without public pricing are estimates.

Best value subscription

Value calculator

TRY YOUR OWN MIX

Set your monthly budget and model mix — every station’s tokens-per-dollar and ranking recompute live. Subscription 6× is the full-quota value; use less and you get less.

Pricing models

PRICING MODELS

Three types: pay-as-you-go (top up, pay per token), subscription (flat monthly), hybrid (tiered + pay-as-you-go). Some stations show prices only after you top up and open the dashboard.

Price details

PRICE PER 1M TOKENS

Per 1M tokens (input / output) in your currency. Baseline = vendor list price (USD converted or CNY direct for Qwen/Kimi). Ratio = station price ÷ baseline. Green = cheaper than official, yellow = slight markup, red = premium, gray = at par. “—” = not listed / login required. Scroll sideways if the table is wide — first column and header stay pinned.

Full directory

FULL DIRECTORY · —

Overseas & official API directory (hand-curated, verifiable domains): 12 vendor-native, 5 cloud-hosted, 16 overseas inference/hosting, 5 overseas aggregators, — total. Overseas-first — no domestic third-party forwarders. ★ = profiled in the value ranking.

SiteDomainRankUptimeModelsScoreNotes / ratioSourceDetails

How the index works

METHODOLOGY

One rule: less money for more tokens means higher value. Other factors answer “is this money well spent?”

① Value per dollar (core)

每元价值(S) = Σᵢ wᵢ × (官方价ᵢ × 汇率) ÷ 该站价ᵢ 最终每元价值 = 每元价值(S) ÷ Σ wᵢ
  • wᵢ Weight for model class i (adjust in calculator; 15 classes: GPT-5.6 ×3, Claude ×3, Grok, DeepSeek ×2, GLM, Qwen, Kimi, Gemini ×3; default is Balanced)
  • Official price Vendor list price from OpenAI / Anthropic / xAI / DeepSeek / Z.AI / Alibaba Cloud / Moonshot / Google (USD converted; Qwen/Kimi are CNY direct), blended 25% input + 75% output
  • Station price Actual station price in your currency, same 25/75 blend
  • Result: how many dollars of official-priced tokens you get per dollar spent. Official = 1.0; below 1 means markup, above 1 means cheaper (discounts/credits).

② Adjustments

  • Coverage C: missing models force you to buy elsewhere → value × C (0–1)
  • Top-up threshold G: higher minimums are less flexible → × (1 − 0.15·min(1, minTopup/200))
  • Credits B: new-user credits, +1 per 20 units, capped at +5
  • Reputation R: track record on uptime and exits (0.5–1.0, 40% weight)

③ Scaled to 100

综合得分 = 每元价值 × C × G × R + B 性价比指数 = 站的得分 ÷ 榜首得分 × 100

④ Limitations

  • Prices change: ranking from —; vendors adjust frequently
  • Cheapest ≠ most reliable: deep discounts often mean new stations or clearance — higher risk of exits, outages or throttling
  • Rate limits (RPM/TPM), concurrency and context caps are rarely published and not scored
  • Reliability and support vary widely — test with the minimum top-up before scaling up
  • Vendor-native stations scoring lower is normal (score = share of coverage); see Price details for single-model truth

About

ABOUT & DISCLAIMER
  • This is an independent price-comparison tool, not an official body, with no affiliation or kickbacks.
  • Sources: each station’s official pricing page plus public rankings/communities (links per station). Collected —. Cross-checked against 2026 public reviews (aiapirank, apiranking, helpaio, baipiao, ai-router-reviews) for direction only.
  • Reputation is a conservative estimate from public info, not a lab test; rankings for stations with hidden prices are indicative only.
  • Third-party platforms are volatile — some may shut down overnight; vendor-native is usually more stable. Test with the minimum top-up first.
  • DIY alternative: register vendor accounts and self-host via one-api/new-api — cost is list price × FX, with the most control and stability, at the price of handling payments and networking.
  • We do not offer top-ups, proxy purchases or acceleration services.