See the full picture
Bring up to six invoices together and compare your effective hourly rates.
CoreForge™ Meter · GPU cost intelligence
More compute.
Less wasted spend.
Understand what your GPUs really cost. Compare up to six invoices with reference rates, flag long-running pods for review, and find a better setup for your workload.
Free report · no account · nothing to install
Private by design. Invoice files are parsed in your browser. Review your costs before choosing to share contact details for a report.
Spot the cost gap
Compare billed compute with reference prices and review potential waste.
Plan your next move
Explore right-sizing, spot instances and inference costs.
Step 1
Add your invoices
CSV or TSV export from RunPod or Vast.ai, an OpenAI or Anthropic usage or cost export, or a RunPod invoice screenshot. Review extracted screenshot charges before adding them. Up to 6 files, so a quarter's worth reads as one picture. GPU invoices and AI API spend are analysed separately.
Supported providers, formats and limitations
| Provider | Accepted files | Billing units | Analysis | Limitations | Pricing source |
|---|---|---|---|---|---|
| RunPod GPU | Invoice CSV/TSV (description, GPU, hours, amount) or an invoice screenshot | GPU-hours; storage, volume and transfer charges | Effective $/GPU-hr per line vs RunPod reference rates; long-running pods; storage share; Vast.ai shown only when cheaper | Credits stay inside compute as unmatched lines; unknown GPUs need manual mapping; USD only | RunPod reference table — live (RunPod API) or built-in, with each quote's datewww.runpod.io |
| Vast.ai GPU | Charges CSV flattened from Vast.ai billing records (type, source, description, amount, item_type, item_description, item_amount; optional gpu_name, num_gpus, label) | Instance-hours × GPUs = GPU-hours (gpu items); disk and bandwidth (bwd/bwu) as storage & transfer | Same GPU-rate comparison as RunPod; payments (credit, transfer, payout, refund) set aside, never counted as spend | Charge records carry no GPU name or GPU count — set them in the mapping step or lines are not compared; Vast.ai rates are indicative marketplace lows that vary by host | Vast.ai indicative 15th-percentile marketplace rate, refreshed every six hours, dated per quotedocs.vast.aidocs.vast.ai |
| OpenAI AI API | Usage export (model, batch, input_tokens, input_cached_tokens, output_tokens) and/or Cost export (line_item, amount_value, amount_currency) | Tokens by type: input, cached input, output; Standard or Batch tier | Spend by model and token type, effective $/1M tokens, billed vs list-price reconciliation, cache and batch shares | Short-context Standard/Batch rates only (no Flex or Priority); models with separate >272K long-context rates are flagged as possibly understated; audio, image, fine-tuning and tool charges stay unpriced; line_item wording parsed heuristically | OpenAI API pricing page, checked 2026-10-16developers.openai.comdevelopers.openai.com |
| Anthropic AI API | Usage report (model, service_tier, uncached_input_tokens, cache_read_input_tokens, cache_creation 5m/1h, output_tokens) and/or Cost report (amount, currency, cost_type, token_type, model, service_tier) | Tokens by type: input, cache read, cache write (5m / 1h), output; Standard or Batch tier | Spend by model and token type, effective $/1M tokens, billed vs list-price reconciliation, cache and batch shares | Cost API amounts are read as cents unless you switch to dollars; fast mode, data residency and Haiku 5.5's prompt-length pricing are not modelled; web search and code execution stay unpriced | Anthropic pricing page, checked 2026-10-16platform.claude.complatform.claude.com |
| Other GPU | Any CSV/TSV with a header row | Whatever you map: description, GPU, hours, amount | Lines whose GPU matches a known class are compared; the rest are listed for review | No usage is invented: unmatched lines are never compared or counted as savings | RunPod reference table |
GPU invoices are compared against GPU reference rates; AI API exports are analysed separately by model and token type and never converted into GPU-hours. Files Meter doesn't recognise go through manual column mapping — nothing is guessed or invented. Amounts in currencies other than USD are set aside, not converted.
Selected files are added automatically. Review their columns below, then run the audit.
Read an invoice screenshot
Choose a PNG, JPG or WebP image from Photos or Files (up to 15 MB). Text recognition runs on your device; the image isn't uploaded. English text and USD amounts are supported. Itemized invoices with GPU models and hours support rate comparisons; monthly billing summaries import as spend only.
Reference rates
What the audit compares against
Refreshes every 6 hours · 11 GPUs · 11 RunPod quotes · Latest check 2026-06-01 00:00 UTC· 11 outdated reference quotes — verify before renting
| GPU | VRAM | RunPod reference | $/hr | Confidence | |
|---|---|---|---|---|---|
| H100 SXM | 80 GB | RunPod | $2.69 | published 2026-04 | |
| H100 PCIe | 80 GB | RunPod | $1.99 | published 2026-04 | |
| H200 | 141 GB | RunPod | $3.59 | indicative 2026-06 | |
| B200 | 180 GB | RunPod | $5.99 | indicative 2026-06 | |
| A100 80GB | 80 GB | RunPod | $1.39 | published 2026-05 | |
| A100 40GB | 40 GB | RunPod | $2.29 | indicative 2026-06 | |
| RTX PRO 6000 | 96 GB | RunPod | $1.79 | indicative 2026-06 | |
| L40S | 48 GB | RunPod | $1.03 | indicative 2026-06 | |
| L40 | 48 GB | RunPod | $0.99 | indicative 2026-06 | |
| RTX A6000 | 48 GB | RunPod | $0.76 | indicative 2026-06 | |
| RTX 4090 | 24 GB | RunPod Community | $0.34 | published 2026-05 |