Look at the jump on DeepSWE
Old result: 7.3 New result: 54.4 Same architecture. Same parameter count. DeepSeek retrained the model and massively improved its performance. That's a huge clue about where AI gains are coming from.
Skywork AI paid version leads in style, presentation, and aesthetic execution.
The real leverage is in the prompt. Export that prompt from DeepSeek or another LLM. Precision in equals excellence out. IO
Deepseek V4 pro quietly closed the gap on claude
This AI model runs nearly nonstop for a fraction of the cost.
$1.02 Spent 923,724 tokens
Deepseek V4 Flash That is all it took to make this small game, I think i might progress this, even further. There are some really rough edges but the type of stuff #opencode go sub can do, is just incredible.
You have reached the end of the archive
All of deepseek