Qwen3.7 Plus is the newest million-token Qwen tier, released June 2026, and it is the value play in the 3.7 line. Qwen3.7 Plus carries the same 1M tokens (1,000,000) context and 65,536 token output as Qwen3.7 Max, at one fifth the input price and 40 percent of the output price: $0.50 per Mtok input and $3.00 per Mtok output. Inside Atlas, reasoning is enabled, so it can be driven through the plan agent before it touches a file.
Why is Qwen3.7 Plus the value play in the Qwen3.7 line?
Qwen3.7 Plus has a 1,000,000 token context and 65,536 token output, identical to Qwen3.7 Max, at $0.50 versus $2.50 per Mtok input. That is one fifth the input price and 40 percent of the output price for the same envelope, released June 2026.
The pricing gap inside the Qwen3.7 line is what makes Qwen3.7 Plus interesting. When two tiers ship the same 1,000,000 token context and the same 65,536 token output, the argument for the expensive one has to be pure capability, and the argument for Qwen3.7 Plus is that most Atlas work does not need the top of any line. Reasoning is enabled on Qwen3.7 Plus, so it can be driven through Atlas's plan agent before it touches a file. Atlas drafts a plan in a read-only plan agent and asks before switching to a build agent, and a reasoning model with a million-token window is exactly what that plan phase wants: enough context to see the problem, enough deliberation to think about it, and a price that lets you run it more than once.
How do you actually fill a 1,000,000 token window in Atlas?
Feed the 1M window on Qwen3.7 Plus through Atlas's tree-sitter AST chunking and RRF-fused search rather than raw file dumps. Atlas indexes code by AST declarations, not blind line windows, so the 1,000,000 tokens fill with complete declarations instead of truncated fragments.
A million tokens is not an invitation to paste. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, which is what RRF-fused search means in practice: a semantic ranking and a keyword ranking, combined so that a result strong on either surfaces. Feeding Qwen3.7 Plus through that pipeline means the 1,000,000 token window fills with the declarations that bear on the task. Raw file dumps do the opposite: they bury the relevant three functions and spend the $0.50 per Mtok input on noise. Atlas can also build its code index with local Ollama embeddings, keeping code off third-party servers, so the retrieval layer that decides what Qwen3.7 Plus sees can run entirely on your machine even though the model itself does not.
What should you never point Qwen3.7 Plus at?
Do not point Qwen3.7 Plus at routine summarization. At $3.00 per Mtok output, Qwen3.7 Plus is 12x Qwen Turbo, and session titles, commit message drafts, and background summaries have no business on a million-token reasoning tier.
The cost mistake with Qwen3.7 Plus is not the input side, it is letting the model touch high-volume, low-value output. $3.00 per Mtok output is 12x Qwen Turbo, and Atlas produces a steady trickle of small generations that nobody reads carefully: titles, summaries, subagent chatter. Set `"small_model": "alibaba/qwen-flash"` at $0.05 per Mtok input so subagents and titles cost almost nothing. Atlas fans out work to subagents that can run in the foreground or in parallel background sessions, and without that small_model pin every one of them bills at Qwen3.7 Plus rates. The other case for switching away from Qwen3.7 Plus is regulatory, covered next.
Can Qwen3.7 Plus run inside a regulated network?
No. Qwen3.7 Plus has closed weights, so there is no local Ollama path for regulated environments. The 1,000,000 token context is only available as a hosted endpoint at Alibaba, and that is a hard constraint no configuration works around.
For teams whose code cannot leave the network, Qwen3.7 Plus is off the table, and it is better to know that before building a workflow around it. Atlas offers a partial hedge: Atlas can build its code index with local Ollama embeddings, keeping code off third-party servers, so the semantic index over your repository can be built and stored locally. But the prompts Atlas assembles from that index and sends to Qwen3.7 Plus do travel to Alibaba. If that is unacceptable, the open-weights Qwen tiers are the path, and you give up the million-token window in exchange for keeping the model in-house. Atlas lets you switch the active model and provider on the fly with favorites and recents, so the same Atlas install serves both postures.
How do you set up Atlas on Qwen3.7 Plus?
Set DASHSCOPE_API_KEY in your environment, or run `atlas login` and select Alibaba, then pin `"model": "alibaba/qwen3.7-plus"` in atlas.json. Confirm the model resolves with `atlas models alibaba` before your first session in 2026.
The full Qwen3.7 Plus setup in Atlas is a credential, a check, and two config lines. DASHSCOPE_API_KEY in the environment, or `atlas login` with Alibaba selected. `atlas models alibaba` to confirm resolution. Then the main model pin, plus `"small_model": "alibaba/qwen-flash"` at $0.05 per Mtok input so subagents and titles cost almost nothing. After that, the operating discipline is retrieval: feed the 1M window through Atlas's tree-sitter AST chunking and RRF-fused search rather than raw file dumps. Every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs, and Atlas computes a unified diff for every file edit and surfaces it for approval before writing, so a newest-checkpoint model still writes nothing you have not seen.
Setup
- 01Set DASHSCOPE_API_KEY in your environment, or run `atlas login` and select Alibaba.
- 02Confirm the model resolves with `atlas models alibaba`.
- 03Pin it as the main model: `"model": "alibaba/qwen3.7-plus"` in atlas.json.
- 04Set `"small_model": "alibaba/qwen-flash"` ($0.05 per Mtok input) so subagents and titles cost almost nothing.
- 05Feed the 1M window through Atlas's tree-sitter AST chunking and RRF-fused search rather than raw file dumps.
Frequently asked questions
- how much does qwen3.7 plus cost
- Qwen3.7 Plus from Alibaba costs $0.50 per Mtok input and $3.00 per Mtok output. That is one fifth the input price and 40 percent of the output price of Qwen3.7 Max, for the same context and output envelope.
- what is the context window of qwen3.7 plus
- Qwen3.7 Plus has a 1M tokens (1,000,000) context window and a 65,536 token output ceiling, identical to Qwen3.7 Max. It is the largest context Alibaba ships.
- qwen3.7 plus vs qwen3.7 max
- Qwen3.7 Plus matches Qwen3.7 Max on context (1,000,000 tokens) and output (65,536 tokens) but costs $0.50 versus $2.50 per Mtok input and $3.00 on output, 40 percent of the Max output price. Qwen3.7 Plus is the value play in the 3.7 line.
- when was qwen3.7 plus released
- Qwen3.7 Plus was released in June 2026 and is the newest million-token Qwen tier, with the largest context Alibaba ships.
- how do i set up qwen3.7 plus in atlas
- Set DASHSCOPE_API_KEY in your environment or run `atlas login` and select Alibaba, confirm the model resolves with `atlas models alibaba`, then pin `"model": "alibaba/qwen3.7-plus"` in atlas.json.
- what small model should i pair with qwen3.7 plus
- Set `"small_model": "alibaba/qwen-flash"` at $0.05 per Mtok input so subagents and titles cost almost nothing. Qwen3.7 Plus charges $3.00 per Mtok output, which is 12x Qwen Turbo, so routine summarization should not touch it.
- can i run qwen3.7 plus locally
- No. Qwen3.7 Plus has closed weights, so there is no local Ollama path for regulated environments. Atlas can still build its code index with local Ollama embeddings, but the model itself is hosted.
- how should i fill a million token context window with code
- Feed the 1M window on Qwen3.7 Plus through Atlas's tree-sitter AST chunking and RRF-fused search rather than raw file dumps. Atlas indexes code by AST declarations and fuses semantic with keyword retrieval by reciprocal rank fusion.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas vs Cursor: terminal AI coding agents compared (2026)
A grounded 2026 comparison of Atlas and Cursor across workflow, change review, extensibility, and pricing for developers choosing an AI coding agent.
Atlas for Groovy: A Terminal-Native AI Coding Agent for Gradle, Spock, and Jenkins in 2026
Atlas is a terminal-native AI coding agent for Groovy in 2026. It reads build.gradle closures and Jenkinsfiles, writes Spock specs, runs ./gradlew test, and applies Spotless.
Atlas for SQL in 2026
Atlas is a terminal-native AI coding agent for SQL in 2026. Run it in a repo with your migrations or .sql files, optimize a query, and review the diff before applying.
Atlas for Fortran: fpm.toml, Explicit Interfaces, and fprettify in 2026
Atlas is a terminal-native AI coding agent for Fortran in 2026. It reads modules, explicit interfaces, and intent declarations, runs fpm test behind a prompt, and runs fprettify.
Atlas vs Base44: Terminal AI Coding Agents in 2026
Compare Atlas, the terminal-native AI coding agent, with Base44, the Wix-owned no-code app builder, for developers in 2026. Evaluate features, pricing, and workflow.
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 JetBrains Junie in 2026: Terminal Agent vs IDE Debugger-Driving Agent
Atlas vs JetBrains Junie in 2026: Junie drives the IDE debugger and reports 61.6% resolved on SWE-Rebench, bundled from $10/mo. Atlas is a free, open source terminal agent with permission gates.
Atlas vs Windsurf: Terminal AI Coding Agents in 2026
Atlas and Windsurf offer distinct AI coding experiences in 2026. Atlas provides a terminal-native TUI with permission-gated tools, while Windsurf features an in-editor live preview and a dedicated GUI.