A rumor has spread that the capabilities of the final version of DeepSeek V4-Pro have leaked. According to investor materials circulating online, its coding ability is only about 0.3% weaker than Claude's flagship model. The materials also said it supports a 1 million-token context and will have a lower API price.
"0.3% Below Claude": What Did the Leak Say?
1. Coding ability: about a 0.3% gap with Claude's flagship, a claim that they are effectively on par.
2. Context: supports up to 1 million tokens.
3. Price: expected to be lower than overseas competitors.
Why the Numbers Can't Be Confirmed: The Source Is Missing
This news is a "leak," not an official announcement. It is based on investor materials circulating online, and DeepSeek has not confirmed them. The figures (0.3%, 1 million tokens, price) are passed-along materials, not an official benchmark.
Don't Mistake a Rumor for Information: Watch These Three Things
1. A sign the open-weights camp's push continues. After last week's Qwen Max open release and Kimi K3 coming to Copilot, this again confirms the trend of DeepSeek pitching frontier-level performance at low cost.
2. But until it's officially confirmed, it is only a "candidate." Don't decide on adoption before real release performance and pricing are confirmed.
3. DeepSeek has recently announced an API price increase, so it is worth watching whether the "cheap frontier" story matches its actual pricing policy.
The Official Announcement Over the Leaked Spec Sheet
The rumor that DeepSeek V4-Pro is coming with Claude-level coding + 1 million tokens + low price hints at the open-weights front's next card. But treat the unconfirmed numbers as reference only and wait for the official announcement.
Related: X post