Changelog · Updated weekly · Last verified 2026-08-23
LLM Pricing Changelog
Every rate change we have recorded across OpenAI, Claude and Gemini, the promotional windows still running, and this site's own release history. Entries are newest first. Rate entries use the provider's effective date; verification entries record our weekly re-check.
Rate changes
Recorded pricing changes
Four changes to the eight models we track. Each was verified against the provider's own documentation before the data file was updated.
2026-08-21 — GPT-5.6 Sol cut to $4.00 / $20.00OpenAI reduced its flagship from $5.00/$30.00 to a promotional $4.00/$20.00 per 1M input/output tokens, a cut of 20% on input and 33% on output. Promotional rate runs through at least November 21, 2026. Verified against OpenAI model documentation on 2026-08-23. See OpenAI API Pricing.
2026-08-16 — Gemini 3.6 Flash promotional pricing confirmedGoogle's standard rate of $1.50/$7.50 per 1M input/output tokens is discounted to $0.75/$3.75 through December 31, 2026 — exactly half. The standard rate resumes January 1, 2027. See Gemini API Pricing.
2026-08-16 — Claude Sonnet 5 price increase cancelledAnthropic's scheduled September 1, 2026 increase to $3.00/$15.00 will not occur; the introductory $2.00/$10.00 rate is now the standard price. Confirmed against Anthropic's official pricing page. See Claude API Pricing.
2026-08-09 — Baseline rate captureFirst verified snapshot of all tracked models: the GPT-5.6 family, Claude Opus 5 and Haiku 4.5, and both Gemini models. Every subsequent entry records a change from this baseline.
Upcoming
Price events still to come
Two promotional windows are currently open. Both carry a published end date, and both would raise costs if they lapse as scheduled — model 2027 budgets on the post-promo number, not today's.
| Date | Model | Promotional rate | Rate if it lapses | Effect on cost |
|---|---|---|---|---|
2026-11-21 On or after | GPT-5.6 Sol | $4.00 / $20.00 | $5.00 / $30.00 | Input +25%, output +50%. On a 1.8B-in / 30M-out month the flagship bill rises from $7,800 to $9,900. |
2027-01-01 From | Gemini 3.6 Flash | $0.75 / $3.75 | $1.50 / $7.50 | Doubles on both axes. On the same workload the bill rises from $1,462.50 to $2,925.00 — still below GPT-5.6 Terra's $3,960.00. |
Promotional end dates are as published by the provider and may be extended or withdrawn; we re-check them weekly and update this table the moment a provider changes the published date. Neither event changes the direction of any comparison on this site — only the size of the gap. Details: OpenAI vs Gemini and Claude vs Gemini.
Verification log
Weekly checks
All eight rates are re-verified against official provider documentation on a weekly cycle. A pass with no change is still logged — "no change" is a finding, not a no-op.
2026-09-14 — Pass, no changesAll 8 models re-checked against official provider documentation; no rate moved. Claude Sonnet 5's cancelled increase to $3/$15 reconfirmed, and the two open price events remain GPT-5.6 Sol on 2026-11-21 and Gemini 3.6 Flash on 2026-12-31.
2026-08-31 — Pass, no changesAll 8 models matched their official sources. GPT-5.6 Sol, Terra and Luna verified at $4.00/$20.00, $2.00/$12.00 and $0.20/$1.20; Claude Opus 5, Sonnet 5 and Haiku 4.5 at $5.00/$25.00, $2.00/$10.00 and $1.00/$5.00; Gemini 3.6 Flash and 3.5 Flash-Lite at $0.75/$3.75 and $0.30/$2.50.
2026-08-23 — OpenAI rates re-verifiedFull re-check of the GPT-5.6 family against OpenAI model documentation, confirming the Sol promotional rate in effect since August 21. This is the verified date now shown on OpenAI rows sitewide.
2026-08-16 — Anthropic and Google rates re-verifiedClaude Sonnet 5's cancelled increase and Gemini 3.6 Flash's promotional window both confirmed against official pages on this date.
CadenceWeekly. Per-model verified dates are printed on every pricing page, and any change lands here with its effective date. See Methodology for sources and process.
Site releases
GetMoreTokens release history
What shipped on the site itself, newest first.
| Date | Release | Pages / files |
|---|---|---|
2026-11-04 | DeepSeek API Pricing | DeepSeek API Pricing: the fourth provider, published as a separate rate card rather than folded into the eight-model weekly panel. Peak rates are deepseek-flash $0.30/$1.20 and deepseek-v4-pro $1.32/$3.96 per 1M tokens, exactly halved off-peak; cache hits cost $0.006 and $0.044 per 1M input (50.0x and 30.0x cheaper than a miss); both models take 1,000,000 input and allow 384,000 output tokens. Search & Fetch billing at $0.75 per 1,000 successful requests began 2026-11-01. Rates read from DeepSeek's official Models & Pricing page on 2026-11-04 and stored in a new window.DEEPSEEK block that cannot be mistaken for the weekly-verified window.MODELS panel; the pricing index basket is unchanged. tools.js gained the deepseek-cost calculator with cache-hit and peak-hour shares. No changes to the eight tracked rates. |
2026-10-28 | LLM Pricing Index | LLM Pricing Index: a fixed basket — 1,000,000 input and 250,000 output tokens on each of the eight tracked models — priced at every checkpoint and indexed to 100.0 at the 2026-08-09 baseline. It reads 87.1 today, down 12.9%: GPT-5.6 Sol's cut is worth −8.7 points and Gemini 3.6 Flash's halving −4.2, while six of eight rates have not moved across ten weekly checks. On the published end dates the index returns to exactly 100.0 on 2027-01-01. Historical and scheduled rate cards were added to the data file as a separate block so they can never be mistaken for live rates. tools.js gained the price-index calculator. No pricing changes. |
2026-10-21 | RAG context cost + agent loop cost tools | Two new calculators for the two most context-heavy request shapes. The RAG Context Cost Calculator prices a retrieved query: 800-token chunks at top-8 send 6,400 retrieved tokens, 15.2x the input of a bare prompt but only 2.7x–3.6x the cost because the answer is unchanged; chunk size and top-k are one dial, and a 10% overlap re-bills 560 tokens a query. The Agent Loop Cost Calculator prices a multi-step loop: an 8-step task sends 66,400 input tokens, 75.9% of it re-read history, and pruning old tool results cuts input by 33.9%. tools.js gained the rag-cost and agent-loop calculators. No pricing changes. |
2026-10-14 | Token pricing explainer + API vs subscription tool | How LLM Token Pricing Works (per-1M unit ladder, the two-counter formula priced on a 3,000-in/600-out request from $0.00132 to $0.0300, standard vs promotional vs long-context rate classes, and the five reasons an invoice misses an estimate) and the API vs Subscription Break-even Calculator (what a flat monthly fee is worth in tokens — 2,500,000 on Claude Opus 5 against 57,142,857 on GPT-5.6 Luna — and the break-even request rate per model). tools.js gained the unit-cost and sub-breakeven calculators. No pricing changes. |
2026-10-07 | Input/output guide + PDF context guide | Input vs Output Tokens (output premiums from 5.00x to 8.33x, the break-even output share that splits the bill, and a live mix calculator) and Can a 200-Page PDF Fit in an LLM? (133,333 tokens, per-model page limits, and the 2,167 tokens of headroom left on Claude Haiku 4.5). tools.js gained the io-split and pdf-fit calculators. No pricing changes. |
2026-09-30 | Cost-reduction hub + conversation growth calculator | How to Reduce LLM API Costs (seven levers ranked by return, an eight-model table showing output as 16.7% of tokens but half the money, and the two promotional expiry dates) and Conversation Cost Growth Calculator, which models quadratic chat-history cost against a sliding window and a rolling summary. tools.js gained the conv-growth calculator with three conversation shapes. No pricing changes. |
2026-09-23 | 1M-tokens guide + prompt weight analyzer | How Many Words Is 1M Tokens? (750,000 words, the request-fit limit, and why OpenAI's 272K surcharge makes a 1M prompt cost more on Sol than on Opus 5) and Prompt Weight Analyzer, which replaces the last "coming soon" card on the homepage. tools.js gained a five-rule prompt analyzer with three measured sample prompts. No pricing changes. |
2026-09-16 | Token definition guide + HTML savings tool | What Is an AI Token? (the cluster root: sub-word definition, conversion by content type, output-to-input multiples) and HTML → Markdown Token Savings, which replaces the "coming soon" card on the homepage. tools.js gained the html-savings calculator with four measured sample pages. |
2026-09-09 | 100K-tokens and context-window guides | How Many Words Is 100K Tokens? and What Is a Context Window?. tools.js gained the context-fill calculator. |
2026-09-02 | Trust pages + Compare cluster complete | About, Methodology, this changelog and Privacy added; trust links added to the footer sitewide. |
2026-09-02 | OpenAI vs Gemini + 1,000-word guide | OpenAI vs Gemini Pricing and How Many Tokens in 1,000 Words?. tools.js gained the words-to-tokens calculator and a long-context preset. |
2026-08-26 | Comparison cluster opens | OpenAI vs Claude and Claude vs Gemini. tools.js gained multi-provider filtering. |
2026-08-19 | Claude and Gemini provider pages | Claude API Pricing and Gemini API Pricing. |
2026-08-11 | Pricing cluster opens | LLM API Pricing Comparison and OpenAI API Pricing. tools.js gained the workload scenario calculator. |
2026-08-09 | Launch | Homepage calculator, Token Counter, API Cost Calculator, Context Window Checker, Tokens ↔ Words, and How Much Does 1 Million AI Tokens Cost?. |
Corrections
Corrections are logged here too.
If we publish a rate, date or calculation that disagrees with a provider's current documentation, we fix the data file and add an entry above — we do not silently edit. That way the record stays auditable and nobody is misled by a number that quietly changed. Report a discrepancy to hello@getmoretokens.com with the page URL and the official source.
Current rate cardLLM API Pricing Comparison.
Input vs outputWhy the reply costs 5.0x–8.3x more per token than the prompt, and the break-even ratio: Input vs Output Tokens.
A 200-page PDF133,333 tokens, and the one model where it nearly does not fit: Can a 200-Page PDF Fit in an LLM?.
Retrieval-shaped costsChunk size, top-k and overlap priced per query, and what one more retrieved chunk costs: RAG Context Cost Calculator.
Agent loopsWhy an 8-step agent sends 66,400 input tokens, 75.9% of it re-read history: Agent Loop Cost Calculator.
Price directionIs token pricing going up or down? The eight-model basket index reads 87.1 against its 2026-08-09 baseline — down 12.9%, with both cuts scheduled to lapse: LLM Pricing Index.
How we verifyMethodology.
Who we areAbout GetMoreTokens.