Compute Price Index
CPIAn equal-weighted reference price for AI compute, derived from 22 models across 9 providers.
1 SCU — STANDARD COMPUTE UNIT
$0.003214
+32.01% · allCost of a reference workload, 1,000 input tokens + 500 output tokens, taken as the equal-weighted geometric mean across the basket. Last reconstituted 5d ago.
Providers in the basket: OpenAI, Anthropic, Google, Alibaba, DeepSeek, xAI, Moonshot, Xiaomi, MiniMax. 22 models, each weighted equally at 1/22 (4.55%).
MODELS
22
one family, one slot, latest version
PROVIDERS
9
equal per-model weight
METHOD
Geometric mean
1/22 (4.55%) each
The basket.
Every model family gets one slot, the latest version always. 22 models total across 9 providers, each weighted equally. Equal-weighted basket v1.0, launched June 18, 2026.
| # | Provider | Model | Released | Input $/1M | Output $/1M | Weight |
|---|---|---|---|---|---|---|
| 1 | Anthropic | Claude Fable 5.1 | 2026-09 | $10 | $50 | 4.55% |
| 2 | OpenAI | GPT-6 Astra | 2026-09 | $10 | $50 | 4.55% |
| 3 | Anthropic | Claude Opus 5.5 | 2026-09 | $4 | $20 | 4.55% |
| 4 | Moonshot | Kimi K3 | 2026-07 | $3 | $15 | 4.55% |
| 5 | OpenAI | GPT-5.6 Terra | 2026-07 | $2 | $12 | 4.55% |
| 6 | Anthropic | Claude Sonnet 5 | 2026-06 | $2 | $10 | 4.55% |
| 7 | OpenAI | GPT-6.1 Sol | 2026-09 | $2 | $10 | 4.55% |
| 8 | Gemini 2.5 Pro | 2025-06 | $1.25 | $10 | 4.55% | |
| 9 | xAI | Grok 4.6 | 2026-09 | $2 | $6 | 4.55% |
| 10 | Alibaba | Qwen3.8-Max | 2026-09 | $2 | $6 | 4.55% |
| 11 | Anthropic | Claude Haiku 4.5 | 2025-10 | $1 | $5 | 4.55% |
| 12 | DeepSeek | V4 Pro | 2026-04 | $1.32 | $3.96 | 4.55% |
| 13 | OpenAI | GPT-5.4 Mini | 2026-03 | $0.75 | $4.5 | 4.55% |
| 14 | Gemini 3.8 Flash | 2026-09 | $0.75 | $3.75 | 4.55% | |
| 15 | Gemini 3.5 Flash-Lite | 2026-09 | $0.3 | $2.5 | 4.55% | |
| 16 | DeepSeek | V4 Flash | 2026-04 | $0.44 | $1.32 | 4.55% |
| 17 | Alibaba | Qwen3.7 Plus | 2026-06 | $0.32 | $1.28 | 4.55% |
| 18 | MiniMax | MiniMax M3 | 2026-06 | $0.3 | $1.2 | 4.55% |
| 19 | Xiaomi | MiMo V2.5 Pro | 2026-04 | $0.435 | $0.87 | 4.55% |
| 20 | OpenAI | GPT-5.4 Nano | 2026-03 | $0.2 | $1.25 | 4.55% |
| 21 | Alibaba | Qwen3.8 Flash | 2026-09 | $0.15 | $0.47 | 4.55% |
| 22 | OpenAI | GPT-6 Luna | 2026-09 | $0.1 | $0.5 | 4.55% |
Retired from the index.
Models that left the basket stay listed with live provider prices — 19 total.
| Provider | Model | Released | Input $/1M | Output $/1M | Retired in |
|---|---|---|---|---|---|
| Alibaba | Qwen3.5 Flash | 2026-02 | $0.1 | $0.4 | v4 |
| Alibaba | Qwen3.6 Flash | 2026-04 | $0.25 | $1.5 | v7 |
| Alibaba | Qwen3.7-Max | 2026-05 | $2.5 | $7.5 | v7 |
| Alibaba | Qwen3.5 Plus | 2026-02 | $0.4 | $2.4 | v4 |
| Anthropic | Claude Fable 5 | 2026-07 | $10 | $50 | v7 |
| Anthropic | Claude Opus 4.8 | 2026-05 | $5 | $25 | v5 |
| Anthropic | Claude Opus 5 | 2026-07 | $5 | $25 | v8 |
| Anthropic | Claude Sonnet 4.6 | 2026-02 | $3 | $15 | v2 |
| Gemini 3.1 Pro | 2026-02 | $2 | $12 | v7 | |
| Gemini 3.5 Flash | 2026-05 | $1.5 | $9 | v7 | |
| Gemini 3.1 Flash-Lite | 2026-03 | $0.25 | $1.5 | v7 | |
| MiniMax | M2.7 | 2026-03 | $0.3 | $1.2 | v4 |
| Moonshot | Kimi K2.7 Code | 2026-06 | $0.95 | $4 | v4 |
| OpenAI | GPT-5.5 | 2026-04 | $5 | $30 | v7 |
| OpenAI | GPT-5.6 Luna | 2026-07 | $0.2 | $1.2 | v8 |
| OpenAI | GPT-5.6 Sol | 2026-07 | $4 | $20 | v8 |
| OpenAI | GPT-6 Sol | 2026-09 | $2 | $10 | v9 |
| xAI | Grok 4.3 | 2026-05 | $1.25 | $2.5 | v4 |
| xAI | Grok 4.5 | 2026-07 | $2 | $6 | v7 |
SCU History
Reconstitution-driven SCU history. Each step is a new on-chain version; values are constant between reconstitutions.
Higher = compute is more expensive.
Current
$0.003214
Period Change
+32.01%
High / Low
$0.003589 / $0.002435
Reconstitutions
9
Display currencies. The canonical SCU is denominated in USD and stored on-chain as USDC microcents. Non-USD values shown above are display-only conversions sourced from a public FX feed. They are not part of the on-chain CPI value and may differ slightly from any specific FX provider.
Compute Power Index
Purchasing power per dollar — higher is better. Anchored to base 100 at the inception baseline.
A reading of 120 means compute is 20% cheaper than baseline. The Compute Power Index is the inverse of the SCU, normalized — when costs go down, the line goes up.
Current Index
75.8
Period Change
-24.2
High / Low
100.0 / 67.8
Baseline
100.0
Reconstitution Log
Every basket update is recorded on-chain with a timestamp, the affected models, and the resulting SCU. The SCU is reconstitution-driven: it is published on-chain when basket composition or provider rate-card prices change.
openai.gpt-sol: gpt-6-sol → gpt-6.1-sol
SCU: $0.003214 → $0.003214(+0.0%)
anthropic.claude-opus: claude-opus-5 → claude-opus-5.5; openai.gpt-luna: gpt-5.6-luna → gpt-6-luna; openai.gpt-sol: gpt-5.6-sol → gpt-6-sol
SCU: $0.003479 → $0.003214(-7.6%)
alibaba.qwen-flash: qwen3.6-flash → qwen3.8-flash; alibaba.qwen-max: qwen-3.7-max → qwen3.8-max; qwen3.7-plus price input -20.0%, output -20.0%; anthropic.claude-fable: claude-fable-5 → claude-fable-5.1; deepseek-v4-flash price input +214.3%, output +371.4%; deepseek-v4-pro price input +203.4%, output +355.2%; google.gemini: gemini-3.1-pro → gemini-2.5-pro; google.gemini-flash: gemini-3.5-flash → gemini-3.8-flash; google.gemini-flash-lite: gemini-3.1-flash-lite → gemini-3.5-flash-lite; openai.gpt: gpt-5.5 → gpt-6-astra; gpt-5.6-sol price input -20.0%, output -33.3%; xai.grok: grok-4.5 → grok-4.6
SCU: $0.003302 → $0.003479(+5.4%)
gpt-5.6-luna price input -80.0%, output -80.0%; gpt-5.6-terra price input -20.0%, output -20.0%
SCU: $0.003589 → $0.003302(-8.0%)
anthropic.claude-opus: claude-opus-4.8 → claude-opus-5
SCU: $0.003589 → $0.003589(+0.0%)
Added gpt-5.6-luna, gpt-5.6-sol, gpt-5.6-terra; alibaba.qwen-flash: qwen-3.5-flash → qwen3.6-flash; alibaba.qwen-plus: qwen-3.5-plus → qwen3.7-plus; minimax.m: minimax-m2.7 → minimax-m3; moonshot.kimi: kimi-k2.7 → kimi-k3; xai.grok: grok-4.3 → grok-4.5; mimo-v2.5-pro price input -56.5%, output -71.0%
SCU: $0.002742 → $0.003589(+30.9%)
Added claude-fable-5; anthropic.claude-sonnet: claude-sonnet-4.6 → claude-sonnet-5
SCU: $0.002435 → $0.002742(+12.6%)
Inception with 18 families
Methodology
The SCU is calculated from public pricing data using a fixed, deterministic formula. Anyone can reproduce it.
Step 1 — Per-model cost
For each model in the basket, compute the cost of the reference workload of 1000 input and 500 output tokens using the provider's published per-token pricing. Prices are USD per 1M tokens.
cost_m = (input_price_m × 0.001) + (output_price_m × 0.0005)Step 2 — Equal-weighted geometric mean
Take the geometric mean of all N model workload costs. Every model is weighted equally. No tiers, no caps, no provider weights. Each model family holds one slot at its latest version, so a provider cannot inflate its representation by shipping extra SKUs.
SCU = (∏ cost_m)^(1/N) for all m in the index, N = 22On-Chain Verification
The basket and SCU live in the OracleRegistry contract on Base. Each version is a single transaction; the on-chain payload is canonical.
API Reference
Public read endpoints. No authentication required. Full reference in the documentation.
MCP server — for AI agents
Query the live oracle from any LLM client. Stdio transport. No API key required.
Install in any MCP client
npx @compute-finance/mcpClaude Code (registers MCP + skills + cost hook)
npx @compute-finance/mcp setupdata · compute · render · analyze · history
Reference: github.com/compute-finance/mcp · /.well-known/mcp/server-card.json