/news
OpenAI slashes Luna by 80% — DeepSeek's cheaper model lands inside Codex six hours later
DeepSeek's lower-cost model beat its own flagship on all nine agent benchmarks and dropped into OpenAI's Codex tooling. The release landed six hours after OpenAI cut Luna by 80%.
DeepSeek released a lower-cost model that, per the original piece, outperforms its own flagship on all nine agent benchmarks in the comparison set. That alone is a notable internal result for a single lab.
The timing is the loud part: it landed inside OpenAI's Codex tooling about six hours after OpenAI announced an 80% price cut on Luna. Whatever the engineering story behind the timing, the optics are a low-cost model appearing in a competitor's developer surface almost immediately after a major price move.
For anyone building agents, the practical question is whether this model behaves well inside the Codex-style tool loop. The original coverage has the benchmark and integration details; the real test will be how it holds up under longer agent trajectories, not just the headline evals.
I am reading this as a signal that the agent-model market is the next place the price/quality arms race is going to play out — not the chatbot tier.