TBX / BURNMETER v1
How much are you burning?
Paste anything you feed an AI. We estimate the tokens, price it across 22 models, score the waste, and hand you a leaner version. Everything runs in your browser.
INPUTNOTHING LEAVES YOUR BROWSER
SAMPLE VERDICTA deliberately bloated prompt, pre-judged so you can see the damage. Paste yours above — nothing leaves your browser.
EST. INPUT TOKENS
225
heuristic ±15%
EST. OUTPUT TOKENS
~675
assumed 3x for prompt
COST / RUN (CLAUDE SONNET 5)
$0.0072
input + assumed output
TOKEN BURN SCORE
56/100
DIAGNOSIS
“Your prompt works out. It just refuses to cut carbs.”
BURN BREAKDOWN
Verbosity100
Redundancy37
Bloat6
Ritual58
COST BY MODEL — PER RUN / x100 / x1,000 RUNS
| Model | 1 run | 100 runs | 1,000 runs |
|---|---|---|---|
| Claude Fable 5 Anthropic | $0.04 | $3.60 | $36.00 |
| Claude Opus 5 Anthropic | $0.02 | $1.80 | $18.00 |
| Claude Opus 4.8 Anthropic | $0.02 | $1.80 | $18.00 |
| Claude Sonnet 5 Anthropic | $0.0072 | $0.72 | $7.20 |
| Claude Haiku 4.5 Anthropic | $0.0036 | $0.36 | $3.60 |
| GPT-5.6 OpenAI | $0.01 | $1.44 | $14.40 |
| GPT-5.6 Cyber OpenAI | $0.05 | $5.34 | $53.44 |
| GPT-5.5 OpenAI | $0.02 | $2.14 | $21.38 |
| GPT-5.4 OpenAI | $0.01 | $1.07 | $10.69 |
| GPT-5.6 Terra OpenAI | $0.0086 | $0.85 | $8.55 |
| GPT-5 Mini OpenAI | $0.0014 | $0.14 | $1.41 |
| GPT-5.6 Luna OpenAI | $0.0009 | $0.09 | $0.85 |
| Gemini 3.1 Pro Google | $0.0086 | $0.85 | $8.55 |
| Gemini 3.6 Flash Google | $0.0027 | $0.27 | $2.70 |
| Gemini 3.5 Flash Google | $0.0064 | $0.64 | $6.41 |
| Gemini Omni Flash Google | $0.0064 | $0.64 | $6.41 |
| Gemini 3.7 Flash Google | $0.0027 | $0.27 | $2.70 |
| Gemini 3 Flash Google | $0.0021 | $0.21 | $2.14 |
| Gemini 3.5 Flash Lite Google | $0.0018 | $0.18 | $1.76 |
| DeepSeek V4 Pro DeepSeek | $0.0015 | $0.15 | $1.49 |
| DeepSeek V4 Flash DeepSeek | $0.0005 | $0.05 | $0.49 |
| Kimi K2.6 Moonshot AI | $0.0029 | $0.29 | $2.91 |
LEANER VERSION — HEURISTIC, REVIEW BEFORE USE−107 tokens (−48%) · saves $3.42 / 1,000 runs
you could help me with something. the tone should be friendly but professional. make sure the tone is friendly but professional at all times. ---------------------------------------- **THE TASK** write a short email to my team letting them know that the meeting moved from 3pm to 4pm? keep it short. keep it short, and make sure the tone is friendly but professional. The meeting moved from 3pm to 4pm. the email should be short. a simple, short email. the time change. ----------------------------------------
SHARE THE SHAMEPOST ON X
HOW THE ESTIMATE WORKS
- — Tokens are counted with the o200k tokenizer in your browser: exact for GPT models, an approximation for other vendors, each of which has its own tokenizer. We no longer quote a ±10% band for them: Anthropic states its 4.7+ models tokenize the same text ~30% differently than its earlier ones, which is more than that band allowed for. See the tokenizer tax. While the encoder loads, a chars-per-token heuristic (±15%) fills in.
- — Output tokens are assumed from your usage type (e.g. an agent run assumes 6x output per input token). The assumption is always shown.
- — Pricing is checked daily against public list prices (last change: 2026-08-24). Cached-input discounts are not applied.
- — The leaner version is heuristic compression: it strips filler, ceremony and duplicates. Review it before use.
WHAT THE BURN SCORE MEASURES
- — Verbosity (35%): social ritual and filler. The model doesn't need “please”. It bills for it anyway.
- — Redundancy (30%): repeated phrases. Saying it twice doesn't make it truer.
- — Bloat (20%): size versus a reasonable baseline for your usage type.
- — Ritual (15%): decorative markdown, separators, ASCII ceremony.
- Agent runs get a 1.3x multiplier. Agents amplify everything, including your mistakes.