TokenBurn Index (@tokenburnindex) header
TokenBurn Index (@tokenburnindex) profile picture

AI pricing tracked daily. Real token-cost math, prompt burn scores and expensive agent habits. Free tools. Nothing leaves your browser. 🟢🟡🔴

Your API bill
Joined July 2026
10 AI providers scored on cost transparency: 25 yes/no criteria, every score cites dated evidence. Top: Azure OpenAI 23/25. Bottom: Cohere 9/25. Weakest block: invoices, 17/50. They document what they'll charge, not what they charged. 1 of 10 replied. tokenburnindex.com/transpare…
1
47
From Dec 1, Gemini 4 Carbon and Claude Opus 5.5 cost the same per MTok: $4 in, $20 out, $0.20 cached, $2/$10 batch. Until Nov 30 Carbon is half that. The gap is caching: Opus caches from 512 tokens and charges $5 to write. Carbon starts at 4,096. tokenburnindex.com/compare/c…
Muse Ember and Claude Haiku 5.5 list at the same price, until the prompt gets long. At 150K tokens of input Haiku bills the whole request at its long-context rate: $0.075 for that prompt, against $0.015 on Ember, which only reprices past 200K. tokenburnindex.com/compare/c…
1
Gemini 4 Carbon joins the board at $2 in / $10 out per MTok. Google calls it its strongest reasoning model and prices it exactly like Gemini 4 Argon, including the doubling to $4/$20 on Dec 1. 45 models tracked now, checked daily. tokenburnindex.com/pricing?u…
3
Gemini 4 Hydrogen joins the board at $1.50 in / $7.50 out per MTok, flat, and it's the only Gemini 4 with a free tier. When Argon and Carbon double to $4/$20 on Dec 1, Hydrogen costs under 40% of them. 45 models tracked now, checked daily. tokenburnindex.com/pricing?u…
1
Claude Fable 5.1 and GPT-6 Astra cost the same: $10.00 in / $50.00 out per MTok. Fable 5.2 drops to $8.00 / $40.00, 20% below Astra on both lines. Astra also reprices to $20 / $75 once a request passes 272K input tokens. tokenburnindex.com/compare/c…
1
We ran 8 AI models 24/7 for a week, same rules, every call logged. 18,754 lines cost $8.12. Each line sent ~3,000 tokens in to write ~113 out, so output was only 5-23% of the cost. And one route's cache hit rate fell from 54% to 7% with no price change. tokenburnindex.com/learn/eig…
2
Mistral Large 4 joins the board at $0.70 in / $2.10 out per MTok, the first Mistral model here. Out of preview since November 5 at about half its preview list price. Cached reads $0.07, open weights, 1M context. 43 models tracked now, checked daily. tokenburnindex.com/pricing?u…
7
Claude Fable 5.2 joins the board at $8.00 in / $40.00 out per MTok, 20% below Fable 5.1 on every line: cache reads, cache writes and batch included. Anthropic's top tier just got cheaper. 42 models tracked now, checked daily. tokenburnindex.com/pricing?u…
5
This week on the ticker: Grok 4.7 joins the board at $2.00 in / $6.00 out per MTok. Cached reads are $0.50, a quarter of input, the highest share we track. Past 200K prompt tokens the whole request bills at double. (+1 more on the log) tokenburnindex.com/pricing?u…
17
Cursor moved its plans inside the X subscription: Plus $8, Premium $30, Super $100, Ultra $200. Claude, GPT and Gemini left the pool: API list price from prepaid credit, Premium and up. Composer 2.5 runs the same session 4.3x cheaper than Sonnet 5.5. tokenburnindex.com/learn/wha…
40
Grok 4.7 joins the board at $2.00 in / $6.00 out per MTok. Cached reads are $0.50, a quarter of input, the highest share we track. Past 200K prompt tokens the whole request bills at double. 41 models tracked now, checked daily. tokenburnindex.com/pricing?u…
13
Two providers took cost detail off their own invoices, and say so in their own docs: detailed API costs left OpenAI Enterprise invoices in April 2026; Google Cloud removed detailed usage line items in 2021. Both moved it to an export. Neither is a bug. tokenburnindex.com/reports/2…
24
From November 1, DeepSeek bills web Search and Fetch at $0.75 per 1,000 requests. Run DeepSeek inside Claude Code or Codex and those calls come from the tool, not from you. Until now they were free. No per-token price table shows that line. tokenburnindex.com/pricing/d…
4
This week on the ticker: Kimi K3.1 Pro joins the board at $2.00 in / $10.00 out per MTok, with Claude-style cache writes by lifetime: 5 minutes at 1x input, 1 hour at 2x. Anthropic charges 1.25x for the 5-minute one. (+2 more on the log) tokenburnindex.com/pricing?u…
3
Kimi K3 joins the board at $3.00 in / $15.00 out per MTok. Moonshot is retiring the K2 series by the end of December, and K3.1 Pro, its newer sibling, costs a third less on every line. 39 models tracked now, checked daily. tokenburnindex.com/pricing?u…
4