LLMPrice indices · 2026-07
Four price indices for AI compute, each pricing the same twelve models on a different workload shape. Base period 2024-07 = 100.
Prices last synced 2026-07-01. This issue is frozen and will not be regenerated.
This issue corrects the basket. Constituents were previously selected from a hardcoded list of model identifiers, which could only ever pick a model already typed into it. As a result the Anthropic flagship slot held claude-opus-4-1 at $15/$75 from December 2025 to July 2026, eight months after Anthropic had shipped cheaper flagship models, overstating that constituent roughly threefold. The xAI, Google and Mistral slots held stale constituents for shorter periods.
Constituents are now selected by product family, so a new release enters by itself and is chain-linked like any other substitution. The previously published figures for this issue were retrieval 76.86, chat 95.28, content 98.53, agent 61.18. They are superseded by the values below. A corrupt archive snapshot for 2024-08-01 was also repaired, adding one reading.
The Flagship Premium definition changed in the same pass. It was a median across all flagships over a median across all values, which compared one provider's flagship to another's value tier. It is now the median of each provider's own flagship-to-value ratio. Both definitions remain in the engine so the revision is checkable.
Added 2026-07-30: the correction restated the entire back series, not only this issue’s readings. The corrected rule selects constituents in every period, so every reading since the base moved: 23 of the 24 previously published readings changed in each of the four profiles, the base period alone unchanged at 100 by construction. The chat profile’s 2025-06 reading, for example, moved from 95.60 to 100.39. Monthly readings rose from 24 to 25 with the snapshot repair, and disclosed substitutions from 11 to 17. This paragraph was added after first publication because the original note disclosed only the four current-issue figures; annotating the issue is the revision path the methodology commits to.
Added 2026-07-31: the Google flagship constituent in this issue, gemini/gemini-3.1-pro-preview, carries a preview-suffixed key while the published inclusion rule excluded previews without qualification. The divergence was found by external audit. The rule as the engine has always applied it admits a preview-named Pro for Google only, because Google sells each new Pro generation under a preview key for months as the current, and only, Pro tier on its own price list; the alternative would freeze the slot on a superseded model. The methodology page now states this exception. The constituent, and every figure in this issue, are unchanged.
Priced on the models a team already runs, AI compute has barely moved in nineteen months. The agent-shaped workload is down 36.0% since 2024-07, but only 5.3% of that came after January 2025.
The frontier did get cheaper over the same period, and by a lot. It got cheaper by replacement: providers ship new models at lower rates far more often than they cut the price of a model already in service. Only 9 of the 24 monthly links in this series carry a price change on a constituent present in both periods. A matched-model index measures the price of what you are already buying, so a saving that arrives as a new model name is invisible to it by construction, and invisible on your bill too until you migrate.
Which is the point of the split below: the content profile now reads 102.67, above its base period, while the cache-heavy agent profile reads 64.02. Whether a team captured any of the decline depended on the shape of its workload, not on which vendor it picked.
Deflation by cacheable share of input
A standing section, reported every issue. The four profiles differ in how much of their input can be served from cache. Ordering the profiles by that share shows whether price movement has tracked it.
| Profile | Cacheable input | Index | Since 2024-07 |
|---|---|---|---|
| Agent Index | 85% | 64.02 | -36.0% |
| Retrieval Index | 70% | 80.24 | -19.8% |
| Chat Index | 30% | 99.47 | -0.5% |
| Content Index | 20% | 102.67 | +2.7% |
Movement is ordered by cacheable share of input this issue: the more of a workload's input can be cached, the further its index has fallen. This is a correlation across four profiles, not a demonstration of intent. It is reported each issue whether or not the ordering holds.
History
The series began 2024-07 and lengthens by one reading each month. 25 monthly readings so far, chat profile shown. It is short, and will stay short for a while.
The LLMPrice 12, as priced this issue
| Constituent | Provider | Line | Cache rate | Chat cost |
|---|---|---|---|---|
| GPT-5.5 gpt-5.5 | OpenAI | flagship | $0.500 | $16.82 |
| GPT-5.4 Mini gpt-5.4-mini | OpenAI | value | $0.075 | $2.52 |
| Claude Opus 4.8 claude-opus-4-8 | Anthropic | flagship | $0.500 | $14.32 |
| Claude Sonnet 5 claude-sonnet-5 | Anthropic | value | $0.300 | $8.59 |
| Gemini 3.1 Pro Preview gemini/gemini-3.1-pro-preview | flagship | $0.200 | $6.73 | |
| Gemini 3.5 Flash gemini/gemini-3.5-flash | value | $0.150 | $5.05 | |
| DeepSeek V4 Pro deepseek-v4-pro | DeepSeek | flagship | $0.004 | $0.588 |
| DeepSeek V4 Flash deepseek-v4-flash | DeepSeek | value | $0.003 | $0.189 |
| Grok 4.3 xai/grok-4.3 | xAI | flagship | $0.200 | $1.72 |
| Grok 3 Mini xai/grok-3-mini | xAI | value | $0.075 | $0.366 |
| Mistral Large (2025-12) mistral/mistral-large-2512 | Mistral | flagship | none published | $1.00 |
| Mistral Medium (2026-04) mistral/mistral-medium-2604 | Mistral | value | none published | $4.50 |
Substitutions
Every basket change since the series began, with the linking factor applied so the change does not appear as a price movement.
| Date | In | Out | Link factor |
|---|---|---|---|
| 2024-08-01 | GPT-4o Mini, Mistral Large (2024-07) | GPT-3.5 Turbo, Mistral Large (2024-02) | 1.0999 |
| 2024-11-01 | Claude Sonnet 3.5 (2024-10) | Claude Sonnet 3.5 (2024-06) | 0.9460 |
| 2025-02-01 | DeepSeek Reasoner, DeepSeek Chat, Grok 2, Mistral Large (2024-11) | Mistral Large (2024-07) | 1.0000 |
| 2025-03-01 | Gemini 2.0 Flash | Gemini 1.5 Flash | 1.0053 |
| 2025-05-01 | GPT-4.1, GPT-4.1 Mini | GPT-4o, GPT-4o Mini | 1.0000 |
| 2025-06-01 | Claude Opus 4 (2025-05), Claude Sonnet 4 (2025-05), Grok 3, Mistral Medium (2025-05) | Claude Opus 3 (2024-02), Claude Sonnet 3.5 (2024-10), Grok 2, Mistral Medium (2023-12) | 1.0000 |
| 2025-07-01 | Gemini 2.5 Pro, Gemini 2.5 Flash, DeepSeek R1, Grok 3 Mini | Gemini 1.5 Pro, Gemini 2.0 Flash, DeepSeek Reasoner | 1.0000 |
| 2025-08-01 | Grok 4 | Grok 3 | 0.9814 |
| 2025-09-01 | GPT-5, GPT-5 Mini, Claude Opus 4.1 | GPT-4.1, GPT-4.1 Mini, Claude Opus 4 (2025-05) | 1.0000 |
| 2025-10-01 | Claude Sonnet 4.5 | Claude Sonnet 4 (2025-05) | 1.0000 |
| 2025-12-01 | GPT-5.1, Gemini 3 Pro Preview | GPT-5, Gemini 2.5 Pro | 0.9993 |
| 2026-01-01 | GPT-5.2, Claude Opus 4.5 | GPT-5.1, Claude Opus 4.1 | 1.0000 |
| 2026-03-01 | Claude Opus 4.6, Claude Sonnet 4.6, Gemini 3.1 Pro Preview | Claude Opus 4.5, Claude Sonnet 4.5, Gemini 3 Pro Preview | 1.0000 |
| 2026-04-01 | GPT-5.4, GPT-5.4 Mini, Mistral Large (2025-12) | GPT-5.2, GPT-5 Mini, Mistral Large (2024-11) | 1.0000 |
| 2026-05-01 | GPT-5.5, Claude Opus 4.7 | GPT-5.4, Claude Opus 4.6 | 1.0000 |
| 2026-06-01 | Claude Opus 4.8, Gemini 3.5 Flash, Grok 4.3 | Claude Opus 4.7, Gemini 2.5 Flash, Grok 4 | 1.0000 |
| 2026-07-01 | Claude Sonnet 5, DeepSeek V4 Pro, DeepSeek V4 Flash, Mistral Medium (2026-04) | Claude Sonnet 4.6, DeepSeek R1, DeepSeek Chat, Mistral Medium (2025-05) | 1.0000 |
Downloads and citation
Series as of this issue, JSON · Series as of this issue, CSV
These files are frozen with the issue and never change when later issues publish, so what a citation of this page resolves to is what was published. Links repointed 2026-07-30; they previously served the living files, which the next issue would have silently changed. The living series is on the latest issue.