Bolt Adds DeepSeek V4.1 Flash at 10x Cheaper Than V4 Pro

DeepSeek V4.1 Flash joins Bolt Forge with cheaper visual iteration

Bolt has added DeepSeek V4.1 Flash to Bolt Forge , its research preview for building apps with selectable open-weight models. Bolt says beta testers preferred the model’s visual output for landing pages and marketing sites, while its lower usage cost allows roughly 10 times as much work as DeepSeek V4 Pro under the same allowance.

The release expands Forge after users spent its first week favoring the fastest available model. V4.1 Flash gives developers a lower-cost option for repeated layout, styling, and content revisions without paying the higher usage rate attached to V4 Pro.

Forge opens the model picker

Bolt Forge exposes experimental open-weight models in the agent picker instead of selecting a model behind the scenes. Its lineup now includes GLM 5.3 Flash, GLM 5.3, Kimi K3, DeepSeek V4 Pro, and DeepSeek V4.1 Flash. Bolt describes Forge as a research preview scheduled to run through mid-October and plans to add new open models as their weights become available.

Bolt advertises up to 50 times more usage for individual Pro subscribers who use Forge, depending on the selected model. Access requires consent to share eligible build-session data with Arcee AI for open-model training.

Flash stretches the iteration budget

DeepSeek V4.1 Flash consumes about one-tenth of the usage charged by V4 Pro, according to Bolt. That difference matters during design-heavy work, where developers may submit dozens of prompts to adjust spacing, typography, color, responsive behavior, and page structure.

A developer can use V4.1 Flash for rapid prototypes, then switch models when a project requires deeper reasoning or ongoing maintenance. The selected model affects future code generation and editing, while the generated application runs independently after deployment.

Each model has a different job

Bolt’s benchmark shows a narrow gap

Forge’s open models scored 92.2 on Bolt’s internal Build Index, compared with 101.0 for the company’s highest-scoring paid closed model. The Forge score equals about 91% of the closed model’s result, although an internal benchmark cannot show how the models will perform across every framework, codebase, or prompt style.

Bolt has not published a separate Build Index score for V4.1 Flash in this announcement. Its quality claim comes from beta-tester feedback that favored the model’s visual output, so developers should compare it with other Forge models on their own components and design systems.

Preview limits require safeguards

Forge remains experimental inside Bolt. The company recommends duplicating a project before switching a serious build into Forge and using Standard or Max for complex production work. Forge also lacks PDF upload support, and model-specific usage rates can reduce the advertised allowance.

• Back up or duplicate a project before changing models.
• Review generated code, dependencies, and configuration before deployment.
• Test responsive layouts and accessibility instead of relying on visual output alone.
• Track usage because the available models consume allowances at different rates.

Training consent covers full sessions

Using Forge’s additional allowance requires agreement to its data-sharing terms. Eligible records can include prompts, generated code, errors, manual edits, retries, and repair traces that show how a build changed during a session.

Bolt presents a consent screen when a user enters Forge. Switching back stops future sharing, while material already incorporated into model training cannot be removed from the resulting model. Bolt says it anonymizes the records and removes secrets, sensitive data, and personal information before transferring them to Arcee AI under a data-processing agreement.

Select V4.1 Flash from the agent picker

Pro subscribers can activate the model by opening the agent picker, selecting Bolt Forge, choosing DeepSeek V4.1 Flash, and sending a prompt. The first Forge prompt triggers the terms and consent flow.

1. Open the agent picker in a Bolt project.
2. Select Bolt Forge .
3. Choose DeepSeek V4.1 Flash .
4. Review and accept the Forge terms.
5. Send a prompt and inspect the resulting changes.

Bolt also offers Bolt Lite , a Forge-only plan priced at $9 per month and aimed at students and side-project developers. Its smaller budget aligns with Flash models designed to support more iterations per allowance.

Frequent releases turn selection into workflow

Bolt launched Forge with four models and added a fifth by the end of its first week. That release cadence lets developers choose among open models based on visual quality, reasoning depth, speed, and usage cost as new weights arrive.

For development teams, the practical change is greater control over model economics. Low-cost models can handle prototypes and repetitive interface edits, while heavier models remain available for tasks that justify their higher usage rates.