InferX Developers @InferXDevs

Let’s Go…. DeepSeek-V4.1-Flash-1017 is live on InferX.net

InferX @InferXai

4 days ago

DeepSeek-V4.1-Flash-1017 is now live on InferX, available at 50% off. Let’s put it to the test and see how it fares against the other DeepSeek Flash models. Try it and tell us what you find. inferx.net

3 2 19 2K 4

421/sec TPS. Devs for the win. 🙌🏼

InferX @InferXai

3 weeks ago

DeepSeek V4 Flash 0731. Still a people’s favorite. 421 tokens/sec. Fast models deserve fast infrastructure. Inferx.net

5 2 37 4K 11

This is what sub-second cold start actually looks like. Real time, no edits.

Everyone should take advantages of these prices. LFG! 🚀

InferX @InferXai

4 weeks ago

Current rates on InferX Some of the lowest inference rates right now: DeepSeek V4 Flash - $0.077 / $0.154 (65% off) DeepSeek V4.1 Flash- $0.12 / $0.48 (60% off) GLM 5.3 Flash- $0.045 / $0.15 (70% off) Input / Output per 1M tokens. Try it out: inferx.net

1 0 5 380 2

Just had a chat with Z AI team. A lot of exciting stuff is cooking.

LFG!! 🚀

InferX @InferXai

a month ago

90% off. For good. We’re making some of our most affordable models even cheaper on InferX: • Qwen3.8 27B FP8 • Qwen3.6 35B A3B FP8 • Qwen3 Coder Next FP8 • Gemma 4 31B IT FP8 • GPT-OSS 20B • Devstral 2 123B 90% off, permanently. No weekend promo. Build away. 🚀

4 0 24 2K 9

@scheemunai Have you tried them on inferx.net

Ask Muse about @InferXai .

InferX @InferXai

a month ago

We asked Meta's Muse for a serverless inference provider with sub-second cold starts, even for large models. Its answer: 'nobody does — except InferX.' Then it explained why — 70B checkpoints are ~140GB, snapshot restore over PCIe takes tens of seconds, most providers can't

3 0 4 638 2

Rogue Claim: The “US companies will serve Kimi K3 at 1/10th the cost” line is being read too literally. If that were already true for buyers, the price would be about $0.30 in / $1.50 out, not $3 / $15.$3 / $15 is the list price. The 10x claim is about unit cost on the GPUs, not the invoice. K3 weights are still ~1.4–1.6TB in MXFP4, so this is not a single-box model. You need at least 8x B300 / GB300 / MI355X, and more once KV cache and concurrency show up. On good US silicon, measured serving cost can fall to cents per million tokens at high utilization. That is the real hardware edge: better chips, better interconnect, fewer cards per replica. Chinese hosts usually need more GPUs and weaker scale-up fabric to run the same model, so same public price does not mean same cost to serve. Utilization wrecks the clean 10x story. A replica at 20–30% utilization does not get the lab-benchmark cost. Idle HBM, bursty traffic, warmup, and long-context cache eat the savings. That is why providers keep the $3 / $15 umbrella even when the raw GPU math looks much cheaper. So the accurate version is: US stacks can generate K3 tokens more cheaply. They have not passed a clean 10x discount to users yet. Rubin could widen the gap. it's not the FLOPs, its Utilization decides whether anyone actually charges $0.30 / $1.50. x.com/rohanpaul_ai/s…

🚀

Guillermo Rauch @rauchg

a month ago

Looks like today may be a record day for token volume % of open models on Vercel AI Gateway: 🟦 Open 78.4% 🟨 Closed 21.6% While spend 💲 usually tells a different story, #3 and #4 today are Moonshot AI & DeepSeek. Adding Z⁠.ai, their combined spend surpasses OpenAI (#2).

238 336 3K 2.8M 818

@rauchg Open-weight models: 56% of tokens, 14% of estimated spend Closed models: 44% of tokens, 86% of estimated spend Anthropic alone: 64% of all spend (and has stayed at 61–65% every month since late 2025)

@ItsmeAjayKV $50? that would be more like $30 on inferx.net.

We Pulled the top models across @OpenRouter @vercel AI Gateway, and InferX over the last 30 days. Same open models. Completely different communities. DeepSeek leads everywhere. But who’s #2 and #3 tells you everything about who’s building where.

We dropped DeepSeek V4.1 Flash pricing by 90% for the weekend. Because inference should be fun n cheaper on weekends too 😎 Have a fun weekend! Inferx.net

GLM 5.3 Flash vs DeepSeek V4.1 Flash on InferX Over the same period: GLM 5.3 Flash • 111% more requests • 70% more input token volume • 94.87% cache hit DeepSeek V4.1 Flash • 94.85% cache hit Users seem to be voting with their workloads.

This is important. At InferX, every model we serve runs on infrastructure we operate, inside our secure container runtime. Our serving infrastructure is in North America unless a customer specifically requests an EU deployment. We don’t quietly route API traffic to third-party inference providers, and we operate with a zero data retention policy. We’ve had plenty of data centers and GPU providers approach us about working together, and we’re open to that. But one thing we can always promise our users is transparency about where their inference is running and how their data is handled.

@InferXai 🤔.....

Interesting signal from OpenRouter: DeepSeek V4 Flash 0731 is still doing nearly 3× the weekly token volume of V4.1 Flash. 13T vs 4.86T tokens. That should tell us something about what users actually want. The newest model doesn’t automatically win. Right performance at the right price matters. Users are voting with their tokens.

Open Models ftw! 🙌🏼

InferX @InferXai

a month ago

We’ll do everything in our power to make sure open-source AI never gets this far. Developers should be free to build. Period.

0 0 4 397 0