DeepSeek: DeepSeek V4.1 Flash 1017 ↗
DeepSeek V4.1 Flash 1017 is the October 17, 2026 snapshot of DeepSeek V4.1 Flash, served by DeepSeek under the `deepseek-flash` API model name. V4.…
Reconnecting the price stream…
LLM token index
What one million AI tokens cost — and how the cost changes.
As of 17 Oct 2026
One million tokens $0.84 Multi-platform basket · seven-day average
Since the previous day 0.9% cheaper Compared with 16 Oct 2026
For 100 million tokens $83.71 An example at the index level
Cost over time
31.7% cheaper over the selected period.
Investera token costSilicon history
Model mix
Recorded usage mix
Priced usage contribution:: OpenRouter 49.9% · Vercel AI Gateway 24.6% · OpenCode 25.6%
Platform selections, not whole-market adoption. The shares measure usage, not users or spending.
Tile area = token weight · colour = model vendor. Hover or focus for exact shares.
Hover the chart. Click to pause.
USD per million tokens. Hover or focus a chart point to see its model mix. 366 daily values, of which 227 are estimates with the most recent known usage mix.
Price-level correlation 0.89 over 241 matching dates. 140 Investera dates use estimated usage. The Silicon history contains chart estimates; its broader market basket differs from these platform selections.
Public per-fixed-date exports and exact provider/model requests give daily usage history without API keys. Older gaps use marked estimates.
The models behind the index
Recorded daily platform usage.
- DeepSeek: DeepSeek V4.1 Flash35.3%
- Z.ai: GLM 5.3 Flash8.4%
- Xiaomi: MiMo-V2.6-Flash5.2%
- Z.ai: GLM 5.34.6%
- Anthropic: Claude Opus 5.54.0%
405 priced models covering 96.4% of blended usage share. The weights sum to 100% across this qualified basket of all priced models with reported usage. DeepSeek and low-cost models are eligible; there is no top-ten cap or input-price floor. They describe the configured platform selections, not the whole AI market. The published price is moreover an average of the previous seven calendar days.
How to read this
- Start with the usage.OpenRouter, Vercel and OpenCode publish dated usage distributions.
- Weight their prices.Qualified models with more token usage weigh more, at a fixed input/output mix.
- Smooth the result.Seven calendar days blend into one average that shows the trend.
The index can fall because prices sink or because usage moves to cheaper models. Your bill depends on model, token mix and discounts.
Sources for 17 Oct 2026
OpenRouter + Vercel + OpenCode usage · Recovered provider/OpenCode prices · Dated price records · Vendor list prices · OpenRouter prices · LiteLLM · Vercel AI Gateway
Recent developments
New listings, price changes and usage updates.
Elorian: Glimpse ↗
Glimpse is a multimodal visual reasoning model from Elorian, built for visual understanding and complex visual reasoning in the physical world. It …
Reflection: Beam 1.0 501B A23B ↗
Beam 1.0 is Reflection's open-weight reasoning model for coding and long-horizon agentic workflows. It uses a sparse mixture-of-experts architectur…
Listings are dates when models were added to OpenRouter. Notes and usage updates come from the watched event records.
All developments & search
Usage sources on 2026-10-17
| Platform | Blending coefficient | Matched model share | Usage date | Priced basket attribution |
|---|---|---|---|---|
| OpenRouter | 50.00% | 100.00% | 2026-10-17 · day | 49.85% |
| Vercel AI Gateway | 25.00% | 63.58% | 2026-10-17 · day | 24.59% |
| OpenCode | 25.00% | 78.51% | 2026-10-17 · day | 25.56% |
The coefficients are renormalised across the available sources. The shares are dimensionless; absolute platform volumes are not summed. Published model shares stay fixed. Unknown names and unpublished tail mass use dated OpenRouter proportions only among models missing from the published list, with the Vercel open/closed totals constraining the remainder, per the Investera model classification. After price exclusions the actual priced attribution can differ from the blending coefficient. Each source can carry previous usage for up to seven days.
Unresolved source model shares
Vercel AI Gateway
- Beam 1.0: 3.12%
- Glyph Cluster: 2.08%
- Jev: 2.58%
- Other: 28.63%
OpenCode
- space-bunny: 11.78%
- beam: 3.84%
- Other: 5.87%
Silicon validation · target 0.90
The target is not yet met. Full shared history: 0.885 (241 dates). Three-source overlap: 0.775 (67 dates). Retrospective holdout: 0.189 (37 dates, 2026-09-10 → 2026-10-16).
Since 2026-07-26: price-level correlation 0.934 over 83 dates; daily-return correlation 0.222 over 82 pairs. 0 index dates carry estimated usage; 0 comparison readings are estimates.
Since 2026-07-26 Investera has changed -36.39% and Silicon -33.79% on shared dates. The largest drawdowns were -49.81% and -35.86% respectively. Mean absolute percentage price error: 11.49%. Correlation alone does not show that the declines match in magnitude.
OpenRouter-only baseline: 0.815. Daily-return correlation for the full history: 0.245 (209 pairs of consecutive days). 140 index days carry estimated usage; 130 comparison readings are chart estimates. Pearson price-level correlation on exact shared dates; daily-return correlation is separate. Estimates and chart-digitized benchmark observations are counted explicitly. The historical holdout is retrospective, not a prospective test of an unseen period. Correlation does not measure price-level accuracy.
Recent validation input: on average 0.00% of the platform coefficients carry previous usage forward. 15.21% of the priced basket weight lacks a saved price observation or effective date; 81.68% has observation-only support. The historical price coverage needs improvement before the correlation target can be considered reliable.
Usage across platforms
Independent selections help explain shifts in model demand. Each platform has its own population and measurement method. Production combines normalised platform shares. The Vercel open/closed distribution governs its contribution; OpenCode provides a model mix. Source coefficients, dates and unresolved coverage are shown next to the selected basket.
Vercel · Open vs. Closed Token Volume ↗
2026-10-17 · 60.8% open weights · 39.2% closed weights
Open means downloadable weights, per the Vercel classification. The line shows daily token share; it does not show request share or spend. Snapshot: 2026-10-18T08:57:56.676Z.
Daily open / closed shares · 62 days
| Date | Open | Closed |
|---|---|---|
| 2026-10-17 | 60.8% | 39.2% |
| 2026-10-16 | 55.0% | 45.0% |
| 2026-10-15 | 49.3% | 50.7% |
| 2026-10-14 | 52.2% | 47.8% |
| 2026-10-13 | 56.2% | 43.8% |
| 2026-10-12 | 57.1% | 42.9% |
| 2026-10-11 | 63.4% | 36.6% |
| 2026-10-10 | 61.3% | 38.7% |
| 2026-10-09 | 55.2% | 44.8% |
| 2026-10-08 | 53.0% | 47.0% |
| 2026-10-07 | 51.5% | 48.5% |
| 2026-10-06 | 50.4% | 49.6% |
| 2026-10-05 | 51.8% | 48.2% |
| 2026-10-04 | 62.7% | 37.3% |
| 2026-10-03 | 57.5% | 42.5% |
| 2026-10-02 | 46.2% | 53.8% |
| 2026-10-01 | 55.8% | 44.2% |
| 2026-09-30 | 61.3% | 38.7% |
| 2026-09-29 | 58.7% | 41.3% |
| 2026-09-28 | 56.7% | 43.3% |
| 2026-09-27 | 75.5% | 24.5% |
| 2026-09-26 | 77.7% | 22.3% |
| 2026-09-25 | 70.1% | 29.9% |
| 2026-09-24 | 73.1% | 26.9% |
| 2026-09-23 | 75.4% | 24.6% |
| 2026-09-22 | 77.1% | 22.9% |
| 2026-09-21 | 76.4% | 23.6% |
| 2026-09-20 | 80.6% | 19.4% |
| 2026-09-19 | 80.1% | 19.9% |
| 2026-09-18 | 78.4% | 21.6% |
| 2026-09-17 | 76.3% | 23.7% |
| 2026-09-16 | 74.9% | 25.1% |
| 2026-09-15 | 74.8% | 25.2% |
| 2026-09-14 | 72.6% | 27.4% |
| 2026-09-13 | 79.5% | 20.5% |
| 2026-09-12 | 78.0% | 22.0% |
| 2026-09-11 | 73.3% | 26.7% |
| 2026-09-10 | 66.4% | 33.6% |
| 2026-09-09 | 56.0% | 44.0% |
| 2026-09-08 | 58.0% | 42.0% |
| 2026-09-07 | 61.8% | 38.2% |
| 2026-09-06 | 65.3% | 34.7% |
| 2026-09-05 | 67.3% | 32.7% |
| 2026-09-04 | 58.9% | 41.1% |
| 2026-09-03 | 60.9% | 39.1% |
| 2026-09-02 | 61.0% | 39.0% |
| 2026-09-01 | 61.3% | 38.7% |
| 2026-08-31 | 61.8% | 38.2% |
| 2026-08-30 | 71.3% | 28.7% |
| 2026-08-29 | 69.9% | 30.1% |
| 2026-08-28 | 62.6% | 37.4% |
| 2026-08-27 | 60.1% | 39.9% |
| 2026-08-26 | 57.6% | 42.4% |
| 2026-08-25 | 52.5% | 47.5% |
| 2026-08-24 | 50.1% | 49.9% |
| 2026-08-23 | 60.7% | 39.3% |
| 2026-08-22 | 61.6% | 38.4% |
| 2026-08-21 | 56.8% | 43.2% |
| 2026-08-20 | 56.6% | 43.4% |
| 2026-08-19 | 56.4% | 43.6% |
| 2026-08-18 | 51.8% | 48.2% |
| 2026-08-17 | 52.4% | 47.6% |
© Vercel, AI Gateway Leaderboard Data · CC BY 4.0
OpenCode · Daily model usage ↗
| Model · 2026-10-17 | Share |
|---|---|
| deepseek-v4.1-flash | 60.8% |
| space-bunny | 11.8% |
| glm-5.3-flash | 6.1% |
| Other | 5.9% |
| mimo-v2.6-flash | 5.7% |
| muse-spark-1.3-contributor | 5.0% |
| beam | 3.8% |
| deepseek-v4-flash | 0.6% |
| nemotron-3-ultra | 0.2% |
| step-5-preview | 0.1% |
The shares include "Other" and sum to 100% within the published daily chart. Source updated: 2026-10-18T08:38:42.000Z. Snapshot: 2026-10-18T08:57:56.676Z. Platform shares are not absolute OpenRouter token counts.
Explore further
Open a section when you want the details.