Deepseek V4 may cost more than flat-rate Codex in real coding use
A personal coding test compared with Gemini, Claude, and Codex after several years away from full-time .NET development. The existing setup used for several AI tools through a workflow. ChatGPT suggested Deepseek could cost less than the current setup, so $20 already available on was used on a real project.
The workflow used Hermes for and Claude for planning, then switched Hermes to Deepseek. The $20 was expected to last one or two weeks, but it ran out in about one hour. Deepseek may be cheaper than Codex for API use, but it can be much worse on cost when compared with .
Coding quality seemed roughly similar to Codex, but the test was too short for a firm judgment.
Key points
- was tested after prior use of Gemini, Claude, and Codex.
- $20 on lasted about one hour, not the expected one or two weeks.
- The workflow used Hermes for and Claude for planning.
- Usage-based API access can be cheaper on paper but costly during long coding work.
- Coding ability seemed close to Codex, but the test was too short to conclude much.