# Databricks AI Index

**Explore the tokenomics of enterprise AI:** which models companies use, where spending goes, and which coding agents deliver the best quality for the cost. Based on aggregated, anonymized data from thousands of enterprises using [Unity Gateway](https://www.databricks.com/product/artificial-intelligence/unity-gateway).

## Token and wallet share by **proprietary** provider

### Unity Gateway Token Share - Proprietary Models

Weekly share of Unity Gateway tokens across proprietary models. Updated Nov 2, 2026.

In the week of Nov 2, the largest token shares were Opus 5.5 (35.8%), GPT 6.1 Sol (14.4%), and Sonnet 5.5 (11.5%). Provider totals: Anthropic 61.5% (80.0% in the week of Jul 20), OpenAI 37.0% (16.4% in the week of Jul 20), and Other proprietary 1.4% (3.7% in the week of Jul 20). Since the week of Jul 20, Opus 5.5 gained the most (0.0% to 35.8%) and Opus 4.8 lost the most (49.8% to 1.3%).

Input plus output tokens routed through Unity Gateway, including traffic routed to external model providers.

![Unity Gateway Token Share - Proprietary Models](https://developers.databricks.com/ai-index/share/token-share/image.png)

[Share page for this chart](https://developers.databricks.com/ai-index/share/token-share)

### Unity Gateway Wallet Share - Proprietary Models

Weekly share of list-price spend across proprietary models on Unity Gateway. Updated Nov 2, 2026.

In the week of Nov 2, the largest shares of wallet were Opus 5.5 (34.7%), Older Anthropic Models (11.4%), and GPT 6.1 Sol (9.5%). Provider totals: Anthropic 70.1% (85.6% in the week of Jul 20), OpenAI 27.4% (10.1% in the week of Jul 20), and Other proprietary 2.7% (4.3% in the week of Jul 20). Since the week of Jul 20, Opus 5.5 gained the most (0.0% to 34.7%) and Opus 4.8 lost the most (53.8% to 3.1%).

Tokens priced at each provider's published list price, with cache reads at the cache-read rate. Includes traffic routed to external model providers; not what Databricks bills.

![Unity Gateway Wallet Share - Proprietary Models](https://developers.databricks.com/ai-index/share/wallet-share/image.png)

[Share page for this chart](https://developers.databricks.com/ai-index/share/wallet-share)

## Token and wallet share by **open-weight** provider

### Unity Gateway Token Share - Open-Weight Models

Weekly share of Unity Gateway tokens across open-weight models. Updated Nov 2, 2026.

In the week of Nov 2, the largest token shares were GLM 5.3 Flash (33.3%), DS V4.1 Flash (28.0%), and GLM 5.3 (21.6%). Provider totals: Zhipu 60.4% (92.3% in the week of Jul 20), DeepSeek 28.2% (0.0% in the week of Jul 20), Moonshot 8.8% (1.4% in the week of Jul 20), and Other open-weight 2.6% (6.3% in the week of Jul 20). Since the week of Jul 20, GLM 5.3 Flash gained the most (0.0% to 33.3%) and GLM 5.2 lost the most (92.3% to 5.4%).

Input plus output tokens routed through Unity Gateway, including traffic routed to external model providers.

![Unity Gateway Token Share - Open-Weight Models](https://developers.databricks.com/ai-index/share/open-source-token-share/image.png)

[Share page for this chart](https://developers.databricks.com/ai-index/share/open-source-token-share)

### Unity Gateway Wallet Share - Open-Weight Models

Weekly share of list-price spend across open-weight models on Unity Gateway. Updated Nov 2, 2026.

In the week of Nov 2, the largest shares of wallet were GLM 5.3 (44.0%), Kimi K3 (21.0%), and GLM 5.2 (11.1%). Provider totals: Zhipu 63.1% (93.8% in the week of Jul 20), Moonshot 23.9% (1.1% in the week of Jul 20), DeepSeek 9.1% (0.0% in the week of Jul 20), and Other open-weight 3.9% (5.1% in the week of Jul 20). Since the week of Jul 20, GLM 5.3 gained the most (0.0% to 44.0%) and GLM 5.2 lost the most (93.8% to 11.1%).

Tokens priced at each provider's published list price, with cache reads at the cache-read rate. Includes traffic routed to external model providers; not what Databricks bills.

![Unity Gateway Wallet Share - Open-Weight Models](https://developers.databricks.com/ai-index/share/open-source-wallet-share/image.png)

[Share page for this chart](https://developers.databricks.com/ai-index/share/open-source-wallet-share)

## How enterprises **mix models** over time

### Companies Using Multiple Model Providers with Unity Gateway

Weekly share of active companies that have used at least two, three or four model providers on Databricks-hosted models. Updated Nov 2, 2026.

In the week of Nov 2: 2+ providers 71.1% (36.0% in the week of Jul 20), 3+ providers 47.9% (12.4% in the week of Jul 20), and 4+ providers 31.4% (4.4% in the week of Jul 20).

Among companies active each week, counts every provider a company has used since July 20, 2026. Gemini and Gemma count as one provider, as do GPT and GPT-OSS.

![Companies Using Multiple Model Providers with Unity Gateway](https://developers.databricks.com/ai-index/share/providers-used/image.png)

[Share page for this chart](https://developers.databricks.com/ai-index/share/providers-used)

### Unity Gateway Token Share - Proprietary vs Open-Weight

Weekly share of tokens on Databricks-hosted models, split between proprietary and open-weight models. Updated Nov 2, 2026.

In the week of Nov 2: Proprietary 69.1% (92.7% in the week of Jul 20) and Open-weight 30.9% (7.3% in the week of Jul 20).

Input plus output tokens to Databricks-hosted models on Unity Gateway. Open-weight: GLM, Kimi, DeepSeek, Qwen, Llama, GPT-OSS and Gemma; the rest is proprietary (Claude, GPT, Gemini and Grok).

![Unity Gateway Token Share - Proprietary vs Open-Weight](https://developers.databricks.com/ai-index/share/open-weight-share/image.png)

[Share page for this chart](https://developers.databricks.com/ai-index/share/open-weight-share)

## What a user spends in a week

### Unity Gateway User Spend Percentiles (Weekly)

Weekly spend at list price for the median Unity Gateway user and the users at the 90th and 99th percentiles. Updated Nov 2, 2026.

In the week of Nov 2, weekly spend per active user was $2.49 (Median), $95.96 (P90), and $699.74 (P99). Compared with the week of Jul 20: Median up 523% from $0.40, P90 up 94% from $49.47, and P99 up 37% from $511.44.

Each user's input, cached and output tokens priced at each provider's published list price; human users only.

![Unity Gateway User Spend Percentiles (Weekly)](https://developers.databricks.com/ai-index/share/spend/image.png)

[Share page for this chart](https://developers.databricks.com/ai-index/share/spend)

## Cost/Quality Frontier for Coding

Measured on an internal benchmark built from Unity Gateway traces of real enterprise coding tasks from the Databricks engineering organization.

### Cost per Task vs Quality

The chart plots aggregate pass rate against average cost per task for 20 models, one point each. Updated Nov 10, 2026.

A model is on the frontier when no other model matches or beats it on both cost and aggregate pass rate. Frontier models from lowest to highest cost:

- GPT 6 Luna (claude code): 89.1% aggregate pass rate, $0.05 per task
- GPT 6.1 Sol (codex): 92.0% aggregate pass rate, $0.76 per task
- DeepSeek V4.1 Flash 1017 (codex): 93.1% aggregate pass rate, $1.02 per task
- Muse Wonder (codex): 94.9% aggregate pass rate, $1.53 per task
- Claude Fable 5.1 (claude code): 96.0% aggregate pass rate, $7.47 per task

Built from internal benchmarks on the Databricks codebase. Each model is shown in the harness, Claude Code or Codex, that sits closest to the frontier.

![Cost per Task vs Quality](https://developers.databricks.com/ai-index/share/scatter/image.png)

[Share page for this chart](https://developers.databricks.com/ai-index/share/scatter)

### Highest Quality

Top 10 models, ranked by highest pass rate on the coding benchmark. The top three are Claude Fable 5.1 (Claude Code, 96.0%), Muse Wonder (Codex, 94.9%), and DeepSeek V4.1 Flash 1017 (Codex, 93.1%).

### Shortest Time per Task

Top 10 models, ranked by shortest average time to finish a task (minutes). The top three are GPT 6.1 Terra (Codex, 3.0 min), Beam-501B-A23B (Codex, 3.8 min), and Muse Wonder (Codex, 5.2 min).

### Lowest Cost per Task

Top 10 models, ranked by lowest average cost per task (USD). The top three are GPT 6 Luna (Claude Code, $0.05), GPT 6.1 Terra (Claude Code, $0.12), and Claude Haiku 5.5 (Claude Code, $0.16).

## Publications

- [The Enterprise Case for Open-Weight Models on Databricks](https://www.databricks.com/blog/enterprise-case-open-weight-models-databricks) (Nov 4, 2026): Why enterprises run a mix of open-weight and frontier models, weighing cost, data control and governance.
- [The new economics of AI cost management](https://www.databricks.com/blog/ai-cost-management) (Nov 1, 2026): An interview on measuring AI by the value it returns per dollar, moving from tokenmaxxing to valuemaxxing.
- [The AI Bill is coming](https://www.databricks.com/blog/ai-bill-coming) (Oct 22, 2026): Why AI coding costs rise with usage, and how engineering leaders control them with gateways, routing and visibility.
- [How to scale agentic applications without creating AI sprawl](https://www.databricks.com/blog/how-scale-agentic-applications-without-creating-ai-sprawl) (Sep 30, 2026): Shared infrastructure for model choice, context and control, so each new agent does not rebuild governance and cost controls.
- [How Databricks rolls out frontier models to 12,000 employees on Day 1](https://www.databricks.com/blog/how-databricks-rolls-out-frontier-models-12000-employees-day-1) (Sep 28, 2026): Databricks' internal process for releasing new frontier models company-wide, covering experimental access, per-user budgets, and evaluation.
- [How we eliminated $1 million a year of wasted AI agent spend in one hour](https://www.databricks.com/blog/how-we-eliminated-1-million-year-wasted-ai-agent-spend-one-hour) (Sep 1, 2026): Using Unity Gateway tracing and Genie One to find and fix bugs in MCP tool servers that caused wasted agent token spend.
- [Smart Routing in Unity Gateway](https://www.databricks.com/blog/smart-routing-unity-ai-gateway-match-frontier-quality-30-lower-cost-task) (Aug 13, 2026): Introduces Smart Routing, a Unity Gateway feature that selects a model for each coding task.
- [Managing AI Coding Costs at Scale](https://www.databricks.com/blog/managing-ai-coding-costs-scale) (Aug 7, 2026): Approaches to controlling AI coding tool spend while keeping broad developer access.
- [Benchmarking Coding Agents on Databricks' Multi-Million Line Codebase](https://www.databricks.com/blog/benchmarking-coding-agents-databricks-multi-million-line-codebase) (Jul 8, 2026): The internal benchmark, built from merged pull requests, behind the cost and quality frontier on this page.
