Ollama Turbo

Original Article ↗ Hacker News Discussion ↗

Ollama has launched Turbo, a $20/month cloud service for running open-weight models such as OpenAI’s new gpt-oss, extending a platform previously known mainly for easy local inference. Commenters welcome the convenience and flat-fee access for development and testing, but question the value versus directly using major providers’ APIs, worry about vague usage limits, and see limited privacy guarantees when everything runs on US-based hardware. A recurring theme is mistrust of VC-backed AI tooling: critics argue that Ollama leans heavily on the llama.cpp/GGML ecosystem while pivoting toward a more traditional SaaS model, raising concerns about governance, future “enshittification,” and over-reliance by other open-source projects.

Partnership, branding, and “local” identity

  • Launch is seen as coordinated with OpenAI via the gpt-oss models; some view this as OpenAI “oss-washing” using Ollama’s reputation.
  • Several commenters are surprised Ollama is an independent, VC-backed company and not part of Meta; some say learning this improves their opinion.
  • Concern that “Ollama” had become synonymous with local, offline use, and this move shifts focus toward being a conventional cloud provider.

Open source, governance, and investor influence

  • Multiple comments argue the real issue is governance, not “open source” per se: without independent foundations, companies can later relicense or restrict (Redis, Elastic, Mongo cited).
  • Ollama is praised for MIT-licensed server code but criticized for being controlled by a single VC-backed company, making long‑term direction and licensing uncertain.
  • Some say investor funding made this kind of monetization inevitable and that people should have expected it.

llama.cpp / ggml attribution and engine debate

  • Strong sentiment that the llama.cpp/ggml author “brought LLMs to the masses” and deserves far more credit and money.
  • Dispute over how much Ollama is “just a wrapper”:
    • Ollama team says they now have their own engine, using ggml for tensors and llama.cpp only for legacy models.
    • Critics reply that ggml is effectively the core of llama.cpp, that differences are small, and accuse Ollama of minimizing this dependence and “gaslighting.”
  • Some users are leaving Ollama for llama.cpp + llama-server, saying it now matches or exceeds Ollama’s usability.

Value proposition and pricing of Turbo

  • $20/month flat fee is compared to ChatGPT/Claude; many want cheaper or purely usage-based options and dislike unspecified “hourly and daily limits.”
  • Supporters see value in:
    • Easy way to test big open models without buying GPUs.
    • A simple, unified local/cloud dev story.
  • Skeptics question why pay $20 for quantized open models when state-of-the-art proprietary models cost the same or less via usage-based APIs.

Privacy, jurisdiction, and data handling

  • “Privacy first” marketing is viewed as under-specified; lack of detailed policies and closed-source desktop app reduce trust.
  • Some see no privacy advantage over other US-based providers; others would pay more for EU/Swiss hosting.
  • Debate over whether US jurisdiction is safer or more dangerous than EU/China; consensus only that local remains best for sensitive data.

Local vs cloud; production vs hobby use

  • Many still see Ollama as an excellent on-ramp: install, download models, and go—especially for less technical users.
  • Some argue it’s mainly a “toy” for individuals, with vLLM/SGLang/Bedrock/Vertex preferred for serious deployments; others say Ollama has benchmarked competitively and can be used in production in constrained environments.
  • Frustration that features like sharded GGUF and Vulkan support lag, with an old Vulkan PR cited as evidence of neglected community contributions.

Community reaction and “enshittification” fears

  • A noticeable split:
    • One camp is angry or wary, seeing a familiar pattern of VC-backed OSS turning into a locked-in, monetized platform (Docker Desktop cited).
    • Another camp defends Ollama: Turbo is optional, core remains open, and projects need revenue to survive; paying for GPUs is framed as fair.
  • Several expect more open, purely local alternatives (llama.cpp, sglang, ramalama, etc.) to benefit if Ollama drifts toward a conventional SaaS model.