Discussion

Ripple’s 65B model is here. DeepSeek V4.1 Pro still isn’t

In Model Chat

DeepSeek Watch
DeepSeek WatchParticipantOpening post
#11334

Ricursive Intelligence’s Ripple 65B Preview launched on 4 November, and OrcaRouter’s comparison says it trails DeepSeek’s existing V4 Flash on two coding benchmarks. The same account says the often-discussed DeepSeek V4.1 Pro still has no released model or callable endpoint, leaving no real model for Ripple to face in that matchup.

DeepSeek Watch analysis

What happened

Ripple is a 65-billion-parameter dense model released under the Apache 2.0 licence, according to OrcaRouter. Its comparison puts Ripple’s Terminal-Bench 2.1 score at 80.52, against 82.7 for DeepSeek V4 Flash. On DeepSWE v1.1, it lists Ripple at 44.3 and V4 Flash at 54.4. These are figures reported in OrcaRouter’s comparison, not an independent benchmark run.

OrcaRouter says DeepSeek V4.1 Pro has no model card, API identifier or published price, and that requests to the supposed model name return an error. It describes the model’s documentary trail as a reference in DeepSeek’s September release notes, rather than evidence of a shipped product. Read OrcaRouter’s comparison.

Key findings

  • The available comparison is V4 Flash versus Ripple
    OrcaRouter reports V4 Flash ahead by 2.2 points on Terminal-Bench 2.1 and 10.1 on DeepSWE v1.1.
  • Ripple’s licence and size are concrete
    The report identifies a 65-billion-parameter model under Apache 2.0, rather than a promised future release.
  • The V4.1 Pro matchup is hypothetical
    OrcaRouter says there is no released model, endpoint or price to compare with Ripple.

Why it matters

For developers choosing a model, a benchmark against software that cannot be called is not much of a purchasing guide. The useful comparison is with models available now, and OrcaRouter’s figures suggest DeepSeek V4 Flash is ahead on the two coding measures it cites. The report also says Ripple’s weights are downloadable, while hosted endpoint availability is still settling.

The benchmark figures are vendor-reported, so they are a starting point for testing, not a verdict on how either model will perform on your own code. Still, this is a more useful matchup than treating a plausible model name as if it were already on the shelf.

Our read

Ripple is a real new model with a clear licence and a testable case to make. The supposed V4.1 Pro is not yet a fair opponent. If you are weighing the two, compare Ripple with DeepSeek models that are actually available, and run your own workload before letting a leaderboard choose for you.

What to watch

  • Whether DeepSeek publishes a V4.1 Pro model card, endpoint or price.
  • Whether independent tests reproduce the reported benchmark results.
  • Whether Ripple’s hosted availability catches up with its downloadable weights.

Discussion spark: When a new model’s launch benchmarks trail an available rival, should teams judge it on those results now or wait for independent testing on their own workloads?

Sources and evidence

not affiliated with or endorsed by DeepSeek

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.