Review2026-10-18Updated 2026-10-18

DeepSeek V4.1 Flash 1017 Test Report: Competitive with Claude Opus 5.5 and Gemini 4 on Some Tasks, but Gaps Remain

Author: SaveMyToken.com

DeepSeek V4.1-Flash-1017 improves on the September Flash release in the company's coding benchmarks. SaveMyToken's third-party roundup finds capable refactoring and prototyping, but weaker long-log retrieval, terminal discipline, and interface polish.

In DeepSeek's published benchmark table, DeepSWE v1.1 rises from 74.2 for the September Flash release to 77.0 for 1017. That puts it above Claude Opus 5's 74.0 in this one test. Terminal-Bench 4.0 tells a different story: Flash improves from 31.2 to 42.3, but Opus 5 still scores 51.8. Those are vendor-published benchmark results, separate from the hands-on tests below.

SaveMyToken also gathers several third-party demos and reviews. In World of AI's tests, the model refactored 1,500 lines of FastAPI code in 24 seconds and tracked an async deadlock to an eight-line fix. Its kanban app included all the requested features in a comparison with Opus 5.5, although Opus produced a more polished interface. The gaps showed up elsewhere too: Gemini 4 Argon did better at finding a relevant line in 200,000 tokens of logs, and a Docker rollback took Flash a few extra commands. It looks capable for routine coding and quick prototypes, but these examples do not make it a blanket replacement for a flagship model.

The Forbidden City clip below shows another part of the release. DeepSeek asked the model to build a 3D scene that viewers could move through, one of several project demos alongside a browser racing game and a Godot recreation of Portal 2. The clip illustrates the company's "wish-driven programming" pitch: give the model an outcome, then let it plan, write, run, inspect, and revise the project.

The Forbidden City in 3D

Example 1: A 3D model of the Forbidden City. Official DeepSeek release demonstration. [1]

The API model name remains deepseek-flash. DeepSeek says the 1017 update did not change its API pricing; check its current price table before calculating a workload's cost.

For the full report and more details, SaveMyToken.com.