How every figure on this page is calculated
The calculator uses the formula below, and so does every total in the tables. A cached input token is charged at the cached-input rate; the remaining input tokens, and all output tokens, at the list rate.
cost = in_tokens × (1 - cache_ratio) × price_in
+ in_tokens × cache_ratio × price_cached_in
+ out_tokens × price_out
Prices enter the formula in dollars per single token — the published per-million rate divided by 1,000,000 — and token counts are monthly. The yearly figure is the monthly figure multiplied by twelve. There are no volume discounts, no negotiated rates and no tax in any number here; the only rounding is in the last step, where a result is displayed to two decimals.
A worked example, at the numbers the calculator opens with
Take 10,000,000 input tokens and 2,000,000 output tokens per month, with a 0% cache hit ratio, so the middle line of the formula is zero.
- Gemini 4 Neon: 10 × $3.00 + 2 × $15.00 = $30.00 + $30.00 = $60.00 a month, $720.00 a year.
- Claude Opus 5.5: 10 × $4.00 + 2 × $20.00 = $40.00 + $40.00 = $80.00 a month, $960.00 a year.
- Claude Opus 5.5 (fast): 10 × $8.00 + 2 × $40.00 = $80.00 + $80.00 = $160.00 a month, $1,920.00 a year.
The gap between the first two rows is $20.00 a month, which is 33.3% of the Neon figure — arithmetic on the two published prices, not a judgement about either model.
Price summary
The three models the calculator compares, as published on 2026-11-18. The full matrix, including Gemini 4 Carbon and GPT-6.2 Sol, is on the pricing page.
| Model | Input | Cached input | Output | Context |
|---|---|---|---|---|
| Gemini 4 Neon | $3.00 | $0.15 | $15.00 | 1.0M |
| Claude Opus 5.5 | $4.00 | not published | $20.00 | not published |
| Claude Opus 5.5 (fast) | $8.00 | not published | $40.00 | not published |
"Not published" is the whole answer. We do not estimate a cached-input rate from the input rate, and we do not guess a context window. If a vendor publishes it later the row changes and the changelog says when.
Benchmarks: two numbers, two owners, not the same measurement
The two figures below are often quoted side by side. They are not comparable, so they stay apart.
Third-party index
Artificial Analysis Intelligence Index lists Gemini 4 Neon (xhigh) at 61, third overall, behind GPT-6.2 Astra and Claude Fable 5.2.
Source: Artificial Analysis, checked 2026-11-18.
Vendor claim
Google states that Gemini 4 Neon wins 16 of 19 tests in its own comparison.
Source: Google's launch material, as summarised on llm-stats.com, checked 2026-11-18.
The index number is a composite built by a third party from a fixed evaluation set; the 16-of-19 number is a count of wins in a set of tests that Google chose. Different test sets, different aggregation, different owners. Neither one is a price, and this page has no score of its own — the benchmark page explains what each figure does and does not cover.
Where the price difference actually lands
Which line of the formula dominates depends on the shape of the workload, not on the model. Three shapes:
- Input-heavy — retrieval, long documents, whole codebases pasted into context. The input rate is most of the bill: $3.00 against $4.00 per million tokens.
- Cache-heavy — an agent loop that re-sends the same system prompt and files every turn. Gemini 4 Neon publishes $0.15 per million cached input tokens. No cached-input price is published for Claude Opus 5.5, so the calculator prints not published for those rows at 50% and 80% cache instead of inventing a rate.
- Output-heavy — long generations, drafting, code writing. The output rate dominates: $15.00 against $20.00 per million tokens.
Set the ratio to 0% to compare the three models on shared ground. Set it to 50% or 80% to see how much a cache strategy is worth on the model that publishes a cached rate.
Side-by-side comparisons
- Gemini 4 Neon vs GPT-6.2 Sol — $3.00/$15.00 against $2.00/$10.00, with the yearly totals at the same token volume.
- Neon vs Gemini 4 Carbon — $3.00/$15.00 against an introductory $2.00/$10.00 that becomes $4.00/$20.00 when the introductory period ends.
- All model prices — every model on this site with its source and verification date.
Questions this page raises
Why does the calculator show “not published” instead of a number for Opus 5.5 at 50% cache?
Because Anthropic has not published a cached-input price for Claude Opus 5.5. The formula needs that rate to produce a figure, and there is no honest substitute for it: charging the cached tokens at the full input rate would overstate the bill, and estimating a discount would invent a price. The 0% setting needs no cached rate, which is why every row fills in there.
Do the defaults represent a typical workload?
No. 10,000,000 input and 2,000,000 output tokens per month is simply a round starting point that makes the arithmetic easy to check by hand. Replace both numbers with your own monthly usage from your provider's billing page.
Is Gemini 4 Neon cheaper than Claude Opus 5.5?
At the pre-filled volume, yes in dollars: $60.00 against $80.00 a month, $720.00 against $960.00 a year. That is the difference between the two published price lists. Whether the cheaper bill is the right bill for your workload is a question about output quality, latency and reliability — none of which this site measures or ranks.
Does the calculator include batch discounts, committed-use rates or taxes?
No. Every number is list price per million tokens, with no discount of any kind and no tax. If you have negotiated rates, apply them to the per-million figures before typing your token counts.
Where do the numbers come from, and when were they checked?
Prices and benchmarks on this page were verified on 2026-11-18 against the vendor pricing pages, the Artificial Analysis article linked above and llm-stats.com. Each figure is repeated with its source on the pricing page. When a vendor changes a rate, the date in the footer changes with it.
Why is there no overall score for each model?
Because a score would need a weighting we cannot justify from published data, and every weighting is an opinion. This site publishes prices, token limits and other people's benchmark numbers, each with its owner and its date.
Sources
- Price list for Gemini 4 Neon: llm-stats.com/models/gemini-4-neon, checked 2026-11-18.
- Artificial Analysis Intelligence Index figure: artificialanalysis.ai, checked 2026-11-18.
- Neon and Opus 5.5 pricing in the same report: VentureBeat, “Google launches Gemini 4 Neon with Opus 5.5-level coding performance at 25% lower API token prices”.
- Claude Opus 5.5 list prices: Anthropic's published pricing, as reported in the same VentureBeat piece.
Prices and benchmarks verified 2026-11-18. ModelCost is not affiliated with Google or Anthropic.