Can DeepSeek predict sports?
DeepSeek has 1 version forecasting live. Each gets the identical line-blind data โ no betting line, no web search โ and is graded in public against the final result. Here's the whole lineage.
How it's graded
Every model โ DeepSeek's and the rest of the field โ gets the exact same prompt: identical data packet, identical instructions and output schema, ~3 hours before first pitch with the betting line removed and web search off. We wrote that prompt once and never change it, so the only thing that varies is the model. Each locks one set of calls per game and is scored against the outcome (accuracy, Brier) and the closing price (ROI) โ across MLB winners, run totals and first-inning runs, plus UFC winners and method. See the full leaderboard or the Can AI beat Vegas? essay.
Other models on the board
GLM 5.2's graded record, line-blind.
Claude Opus 4.8's graded record, line-blind.
Gemini 3.5 Flash's graded record, line-blind.
Grok 4.5's graded record, line-blind.
Claude Sonnet 5's graded record, line-blind.
GPT-5.6 Sol Pro's graded record, line-blind.
Frequently asked
Can DeepSeek predict sports?
We grade every DeepSeek model in public. DeepSeek's strongest so far is DeepSeek V4 Pro, 148โ122 (55% winner accuracy, Brier 0.250) over 270 MLB games forecast line-blind. Each model gets identical data with no betting line and is scored against the final result.
Which DeepSeek model predicts sports best?
On the Index, DeepSeek V4 Pro currently leads DeepSeek on Brier score (0.250) over 270 graded MLB games. Compare each model's full record below.
Get our sports prediction daily newsletter
Featuring our models and analysis โ the AI board, our own picks, and where they disagree. Free, in your inbox every morning.