Databricks Foundation Model APIs serve Claude, GPT-5, and Gemini from inside the lakehouse, which means your Unity Catalog governance, audit trail, and network boundary apply to the model call the same way they apply to a table read. Inside Atlas, the terminal-native coding agent, Databricks Foundation Model APIs are the pick for regulated teams. Pricing runs Claude Opus 4.7 $5.00/$25.00 per Mtok, Claude Sonnet 4.6 $3.00/$15.00, GPT-5.4 $2.50/$15.00, Gemini 3 Flash $0.50/$3.00, GPT OSS 120B $0.072/$0.28, and GPT-5 Nano $0.05/$0.40, on top of DBU consumption.
What are Databricks Foundation Model APIs best at inside Atlas?
Databricks Foundation Model APIs are best inside Atlas for governance. Claude Opus 4.7 at a 1,000,000 token context and GPT-5.5 at 1,050,000 sit behind the same DATABRICKS_TOKEN as your data, so an Atlas agent that reads production schemas never leaves the governance boundary.
That is the entire pitch, and for regulated teams it is enough. Unity Catalog governance, the audit trail, and the network boundary all apply to a model call the same way they apply to a table read. Atlas already permission-gates every tool call against allow, ask, and deny rules before it runs, and it computes a unified diff for every file edit and surfaces it for approval before writing. Layering Databricks Foundation Model APIs underneath means the model itself is inside the same boundary as the data the agent is reasoning about, not just the tool calls. For a team that cannot send production schema context to a public API, that is the difference between running Atlas and not running Atlas.
Which models do Databricks Foundation Model APIs carry?
The Databricks Foundation Model APIs catalog spans all three frontier labs. Anthropic, OpenAI, and Google are all present, plus GLM-5.2 and Kimi K2.7 Code, so model switching inside Atlas does not mean a new vendor contract in 2026.
That breadth is the second reason regulated teams pick Databricks Foundation Model APIs. Atlas lets you switch the active model and provider on the fly with favorites and recents, and on a normal setup switching from Claude Sonnet 4.6 to GPT-5.4 means a second vendor, a second key, and a second security review. On Databricks Foundation Model APIs it means changing a string in atlas.json. Claude Opus 4.7 is $5.00/$25.00 per Mtok, Claude Sonnet 4.6 is $3.00/$15.00, GPT-5.4 is $2.50/$15.00, and Gemini 3 Flash is $0.50/$3.00, all behind one DATABRICKS_TOKEN.
How much do Databricks Foundation Model APIs cost?
Databricks Foundation Model APIs token prices sit on top of DBU consumption, so the per-Mtok number is not your all-in cost the way it is on DeepInfra. Claude Opus 4.7 is $5.00/$25.00 per Mtok and Claude Sonnet 4.6 is $3.00/$15.00, plus DBUs.
That layering is the honest tradeoff of an enterprise platform. You cannot forecast an Atlas agent run from the model list alone, because DBU consumption is a second axis. What is genuinely surprising is that GPT OSS 120B is $0.072/$0.28 per Mtok on Databricks Foundation Model APIs, half the Groq price ($0.15/$0.60), which is unusual for an enterprise platform. GPT-5 Nano at $0.05/$0.40 is the other cheap row. Both make good `small_model` pins, so Atlas title, summary, and subagent overhead does not run at Claude Opus 4.7's $25.00 output rate.
What context window do Databricks Foundation Model APIs give Atlas?
Databricks Foundation Model APIs reach 1,050,000 tokens on GPT-5.5 and GPT-5.4, and 1,000,000 tokens on Claude Opus 4.7 and Claude Sonnet 4.6. A million-token window on a governed endpoint is the rare combination that makes Databricks Foundation Model APIs worth the DBU layer.
For an Atlas build session on a large codebase, the 1,000,000 token Claude Sonnet 4.6 row at $3.00/$15.00 per Mtok is the documented coding pin. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion and indexes code by AST declarations using tree-sitter, not blind line windows, so it does not need a million tokens to find the right function. But a long agentic session accumulates plans, tool output, and unified diffs, and a 1,000,000 token window means that accumulation does not force a context reset partway through a refactor.
When should I pick a different provider than Databricks Foundation Model APIs?
Pick a different provider than Databricks Foundation Model APIs if you are not already a Databricks customer. Databricks Foundation Model APIs require a Databricks workspace, and the base URL is literally templated with your host (https://${DATABRICKS_HOST}/ai-gateway/mlflow/v1), which is dead weight otherwise.
There is no scenario where a developer with no Databricks account should provision one just to run Atlas. Claude Sonnet 4.6 at $3.00/$15.00 per Mtok is available first-party without the DBU layer, and without a workspace hostname to template. Databricks Foundation Model APIs earn their place when the governance boundary is the requirement: your data is in the lakehouse, your Unity Catalog policies are the control plane, and an Atlas agent reading production schemas needs to stay inside them. If that describes you, the DBU layer is a cost you are already paying. If it does not, it is pure overhead.
Setup
- 01Export DATABRICKS_HOST (your workspace hostname) and DATABRICKS_TOKEN (a personal access token).
- 02Run `atlas models databricks` to confirm both env vars resolved and the catalog loaded. The base URL is templated: https://${DATABRICKS_HOST}/ai-gateway/mlflow/v1
- 03Pin `"model": "databricks/databricks-claude-sonnet-4-6"` in atlas.json for the 1M-context coding slot at $3.00/$15.00 per Mtok.
- 04Set `"small_model": "databricks/databricks-gpt-5-nano"` at $0.05/$0.40 per Mtok for the cheap slot.
- 05If you want an open-weights cheap slot instead, use `"databricks/databricks-gpt-oss-120b"` at $0.072/$0.28 per Mtok.
Frequently asked questions
- how to use databricks foundation model apis with atlas
- Export DATABRICKS_HOST (your workspace hostname) and DATABRICKS_TOKEN (a personal access token), run `atlas models databricks` to confirm both resolved, then pin `"model": "databricks/databricks-claude-sonnet-4-6"` in atlas.json for the 1M-context coding slot.
- can an ai coding agent use claude inside a governance boundary
- Yes. Databricks Foundation Model APIs serve Claude Opus 4.7 at a 1,000,000 token context behind the same DATABRICKS_TOKEN as your data, so Unity Catalog governance, the audit trail, and the network boundary apply to the model call the same way they apply to a table read.
- how much do databricks foundation model apis cost per million tokens
- Claude Opus 4.7 is $5.00/$25.00 per Mtok, Claude Sonnet 4.6 is $3.00/$15.00, GPT-5.4 is $2.50/$15.00, Gemini 3 Flash is $0.50/$3.00, GPT OSS 120B is $0.072/$0.28, and GPT-5 Nano is $0.05/$0.40. Those sit on top of DBU consumption.
- what context window do databricks models support
- Databricks Foundation Model APIs reach 1,050,000 tokens on GPT-5.5 and GPT-5.4, and 1,000,000 tokens on Claude Opus 4.7 and Claude Sonnet 4.6. The 1,000,000 token Claude Sonnet 4.6 row is the documented Atlas coding pin.
- do i need a databricks workspace to use databricks foundation model apis
- Yes. The base URL is templated with your host (https://${DATABRICKS_HOST}/ai-gateway/mlflow/v1), so Databricks Foundation Model APIs are dead weight if you are not already a Databricks customer. Use a first-party provider instead.
- is gpt oss 120b cheaper on databricks than groq
- Yes. GPT OSS 120B is $0.072/$0.28 per Mtok on Databricks Foundation Model APIs, half the Groq price of $0.15/$0.60. That is unusual for an enterprise platform and makes it a good open-weights small_model pin.
- what small model should i pair with claude on databricks
- Set `"small_model": "databricks/databricks-gpt-5-nano"` at $0.05/$0.40 per Mtok, or `"databricks/databricks-gpt-oss-120b"` at $0.072/$0.28 for an open-weights cheap slot. Either keeps title and summary calls off Claude Opus 4.7's $25.00 output rate.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Audit a Repo with Parallel Subagents in Atlas (2026 Workflow)
How to audit a repo with parallel subagents in Atlas in 2026: the task tool launches explore subagents in their own sessions, so only conclusions return to your context.
Atlas vs Gemini CLI: A Developer's Guide to Terminal AI Agents in 2026
Atlas vs Gemini CLI in 2026: Compare terminal AI coding agents. Atlas offers permission-gated tool calls and diff-based approvals. Gemini CLI provides a million-plus token context and a free tier.
Atlas vs Sourcegraph Cody: Terminal AI Coding Agents in 2026
Comparing Atlas, the terminal-native AI coding agent, with Sourcegraph Cody, an editor extension for whole-codebase search, for developers in 2026. Explore features, pricing, and workflow.
Atlas vs Mistral Vibe for Code: Terminal AI Coding Agents in 2026
Compare Atlas and Mistral Vibe for Code in 2026. Atlas offers terminal-native TUI, explicit diffs, and BYO models. Mistral Vibe for Code provides a four-model stack, multi-platform access, and EU data sovereignty.
Atlas vs Magic.dev: Terminal AI Coding Agents in 2026
Compare Atlas, the terminal-native AI coding agent with permission-gated tools and diff review, against Magic.dev's research claims of 100 million token context models in 2026.
Atlas for Java in 2026
Adopt Atlas, the terminal-native AI coding agent, for Java development in 2026. Enhance your workflow with intelligent code search, refactoring, and robust safety features for Maven and Gradle projects.
Atlas for Axum in 2026
Atlas is a terminal-native AI coding agent for Axum in 2026. It decodes tower trait-bound errors, adds IntoResponse types, and runs cargo nextest run.
Atlas vs Amp: Terminal AI Coding Agents in 2026
Compare Atlas, a terminal-native AI coding agent with free core and local embeddings, against Amp, Sourcegraph's agent featuring Oracle and Orbs, for developers in 2026.