On: GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price

[taylorfinley]

Were you using OpenRouter? I've used 1.8bn tokens in the past week from DeepSeek themselves and 99.2% were cache hits. Total cost was $18.13 usd.

[faitswulff → taylorfinley]

For readers wondering, OpenRouter isn’t capable of caching as effectively as DeepSeek is because they will, for instance, switch inference providers in the middle of a session.

[Computer0 → faitswulff]

The way I use openrouter is I find a model/provider combination I like then pin all requests for that model to that single provider.

[swingboy → Computer0]

If you disable all other providers but DeepSeek in your OpenRouter guardrails, is that effectively the same thing?

[hyprwave → swingboy]

Sure but you still pay only OpenRouter :)

[faitswulff → hyprwave]

Interestingly, OpenRouter will serve inference at cost for a lot of models.

[bogdan → faitswulff]

No? They charge 5.5% (I believe) on every top-up. Which is the equivalent of charging you 5.5% on every request.

[faitswulff → bogdan]

Ah, so that's where the profit comes from. Very payment-processor like move, makes sense that Stripe picked them up.

[roarkeful → Computer0]

How do you do this?

[pests → roarkeful]

provider: { order: ['deepinfra/turbo'], allowFallbacks: false, },

https://openrouter.ai/docs/guides/routing/provider-selection

[nextaccountic → pests]

What harnesses are capable of doing this provider pinning when doing requests on openrouter? Can DeepSeek harness (dsh) do it? Or opencode etc

I stopped using opencode because it has some issues with caching, so I suppose it doesn't do this

[pests → nextaccountic]

Deepseek plugins* all the way down. It should be easy to add support for this.

Using this interesting framework called Cordis I've recent discovered.

[hetspookjee → nextaccountic]

Open code and pi can do it. Unironically, I let Claude configure these things for me.

[pwython → roarkeful]

What pests said. And you can make a preset and pass "model": "@preset/deep-seek"

[gigatexal → faitswulff]

How can they? If you use openrouter Deepseek fine but can’t you select the model direct from Deepseek you’re basically getting api access models that way.

[rapind → taylorfinley]

Fireworks directly. At the time they were the best value of cost, speed, ZDR. They got slower on me though, but I think they are retooling, so maybe things have or will get better again. I think fireworks is primarily for when you want to do your own training on top, which I wasn't doing.

[ampdot → rapind]

DeepSeek, DigitalOcean, GMICloud, and NovitaAI were the only OpenRouter providers that didn't lead to major performance degradation for me