Developer Rewards: DeepSeek V4 at 63% of list

SafeRouter Developer Rewards is open. Do three things and your account rate on DeepSeek V4 drops to 63% of list price, and it holds there for up to 180 days. Nothing about the model changes, nothing about the guarantees changes, and there is no minimum spend to qualify.

The list price on SafeRouter is DeepSeek's own published rate passed straight through, $0.14 per million input tokens and $0.28 per million output. At 63% that becomes $0.0882 and $0.1764. Cached input, already the cheapest line on the bill at $0.0028 per million, comes down to $0.001764.

Why this model, and why now

DeepSeek V4 arrived in April as open weights under an MIT license, two models in the family: V4-Pro at 1.6 trillion parameters and V4-Flash at 284 billion, both carrying a 1-million-token context window over the API. General availability followed on July 20. In the four months since, it has become one of the most talked-about releases of the year among developers, and the reason is specific rather than general enthusiasm.

The interesting part is the agent numbers. DeepSeek reports V4-Flash beating V4-Pro-Preview outright on nine separate agent benchmarks, including 82.7 on Terminal Bench 2.1 for long-horizon shell work and 70.3 on Toolathlon for multi-tool orchestration. Independent measurement from Artificial Analysis puts Flash's Intelligence Index at 52, third among the 101 models it tracks, against a field median of 26. For a model at this price, that combination is unusual enough that a lot of teams have spent the summer rebuilding their agent loops around it.

The honest other half: on public coding leaderboards the V4 family currently sits behind GLM 5.1 and Kimi K2.6, and further behind Western frontier models like Claude Opus 4.8. This is not a frontier-class model and we are not going to pretend otherwise. It is a genuinely capable model at a price roughly two orders of magnitude below the top of the market, which makes it the right default for volume work and exploration, and the wrong choice for the last review pass on something that matters. We wrote that comparison up in detail when the model went live, including the cost-normalized case for running it three times instead of a frontier model once.

How to join

1. Create a SafeRouter account, then open the Referral page and copy your personal invite link.
2. Quote our campaign post on X and add your invite link to the quote.
3. Submit the form with your account and a link to your post.

Once we have reviewed the entry, the rate is applied to your account directly. There is no code to enter and no separate step to claim it. Open the Discount tab in account settings and it will be listed there alongside anything else applying to your account.

The invite link is the part worth paying attention to after you have joined. Developers who sign up through your link are attributed to you, and that attribution is what unlocks the higher reward tiers.

How this combines with platform pricing

Your Developer Rewards rate is an account rate, which means it multiplies with any platform-wide rate the model happens to be carrying rather than replacing it. If we put a promotional platform rate on V4 Flash while your account rate is active, you get both, and the price you see on the models page is already the combined figure. We explained the mechanics of that when platform discounts shipped, and the same arithmetic applies here.

What does not change

Every call still runs inside the hardware-encrypted enclave, still gets written into the transparency log, not sampled, all of them, and still comes back with proof headers you can verify independently against that log. A promotional rate does not buy a lower standard of evidence. The cheapest model in the catalog on the deepest discount we offer carries exactly the same proof as the most expensive one at list price.