I have been testing DeepSeek-V4-Pro with the Pi coding agent. I am mindblown by how well it works out of the box. A few notes: I spent a few hours building an LLM wiki with an agent powered entirely by DeepSeek-V4-Pro on Fireworks inference. This is the first time... I feel like there is an open-weight model that can reason at the level of Claude and Codex. And it does this in a cost-effective way with support for 1M context length. To be clear, I am using DeepSeek-V4-Pro inside of Pi without any special configuration. It works out of the box. It's exciting that there is a model that can just be plugged into a basic harness like Pi, and it just works. I've never seen that before. Most models require lots of configuration and setup. DeepSeek's DeepSeek-V4-Pro is clearly good at agentic coding (probably the best from the open-weight models), but the model is also great on knowledge-intensive tasks where reasoning matters. The agent pulled agentic engineering best practices from different company docs (Anthropic, OpenAI, Google, Stripe, Meta, Modal, DeepSeek, Mistral, Cohere), searched and digested Reddit and HN threads, summarized arxiv papers, and surfaced trending GitHub repos. Then it distilled everything into actionable tips across categories. I love the Wiki it built. The quality is really good. Here is a snapshot of what the wiki looks like: DeepSeek-V4-Pro handled the task without breaking stride. Multi-step research queries, code generation for scaffolding, context-heavy reasoning across disparate sources. For coding specifically, this is the first open-weight model that genuinely feels like a Codex or Claude Code experience. It compares in capability and actual multi-turn agentic work. What made the loop feel so responsive was Fireworks' inference speed (the fastest in the market) and the fact that they actually validate models at the systems level before shipping. No corrupted reasoning traces. Just fast, reliable iteration. The hybrid CSA and HCA attention design cuts KV cache to just 10% and inference FLOPs by nearly 4x at 1M-token context. This is what makes the agent loop actually fast and cheap enough to run in practice. For devs who've been watching open-weight models close the gap but haven't found one that actually delivers in practice, this is the closest I've seen. Try it here:show more

Spent a few hours on V4-Pro last week and the price-performance jump on reasoning-heavy tasks is real. The catch most teams miss is data residency. For Indian and EU enterprise buyers, the model has to be hosted somewhere they trust before any of the cost gains matter. Capability is not the bottleneck for them, hosting is.

@FireworksAI_HQ I have been using it for a day with @openclaw; I'm stoked to have something this good and cheap. Still testing, can't wait to try mimo.

@FireworksAI_HQ are you using v4 pro high thinking or max thinking or no thinking?

@FireworksAI_HQ i set the default to medium, and it works great like that. i haven't had the need to push it to max thinking yet

@FireworksAI_HQ Deepseek was the 2nd best Ai 6 months ago, the competition got way bigger now, but still deepseek initial training is better then most new models which makes it capable of flipping Mythos in a few months

@FireworksAI_HQ Have you tested MiMo-V2.5-Pro?

@FireworksAI_HQ I haven't. Is this available in the Fireworks API?

@FireworksAI_HQ Doesn't seem like it based on Get on that @FireworksAI_HQ

@garrytan @FireworksAI_HQ Pi versus hermes, any experience?

@garrytan @FireworksAI_HQ haven't tested against Hermes, but that could be interesting. i like the simplicity of Pi

@FireworksAI_HQ got this cool harness for my deepseek called mozilla firefox works about the same

@FireworksAI_HQ how many tokens did you burned and how much you were billed. jst wnt to do cost comparison too

@FireworksAI_HQ How are you using it in pi, exactly? What provider?

@garrytan @FireworksAI_HQ I used their flash model through open router and they did amazing for my agentic loop. I’m gonna switch it over to pro version, very excited after seeing this.

@FireworksAI_HQ Cool, I will have to try them together. What do you use for an AST for your codebase and memory?

@FireworksAI_HQ Whoa, DeepSeek-V4-Pro on Fireworks for a full LLM wiki? That’s wild—curious how much custom scaffolding you needed, or did it really just work? 😅

@FireworksAI_HQ Great one

@FireworksAI_HQ KV cache 压到 10%，1M token 下 FLOPs 砍近 4x 大概是长上下文 agent的关键 不过 "out of the box" 我保留意见。Pi 本身就是很厚的 harness，模型插进去能用，一半功劳在 Pi 的 tool routing 上。

@FireworksAI_HQ How did it handle ambiguous requirements where you'd normally need back-and-forth with a frontier model? That's usually where non-frontier falls apart.

@FireworksAI_HQ Wiki generation is the real stress test. Structured output. Agentic iteration on state. If DeepSeek handles this, it handles anything. Need to try

@FireworksAI_HQ Cool thing happened when I used it:

@FireworksAI_HQ that sounds wild! results been consistent so far?

@FireworksAI_HQ Deepseek v4 Pro has been really impressive with Hermes as well

@garrytan @FireworksAI_HQ Better than kimi k2.6 and GLM5.1?

@FireworksAI_HQ Good to see open source models finally catching up. I

@FireworksAI_HQ been waiting for an open-weight model that just works without 3 hours of config 😭 sounds like this might finally be it

@FireworksAI_HQ If you’re a developer who lives in the terminal, this might be worth checking out:

@garrytan @FireworksAI_HQ Pi with glm5.1 on coding plan

@FireworksAI_HQ First open-weight model that actually feels like Claude Code in a real agentic loop is a significant milestone. The gap was supposed to take longer to close

@FireworksAI_HQ An agent-built LLM wiki running entirely on DeepSeek-V4-Pro working well enough to recommend is the milestone. What almost broke the pipeline?

@FireworksAI_HQ shouting this out in today's InnoFlow brief!!

@FireworksAI_HQ It reminds me of how the trad models used to work before they nerfed the hell out of them

@FireworksAI_HQ Agentic coding on open weights finally feels like a real alternative—curious how it holds up on long-horizon refactors with messy legacy code, not just greenfield builds.

@FireworksAI_HQ @grok what's the GUI app in the video

@FireworksAI_HQ Great to hear about your success, @omarsar0! Can't wait to see what amazing things you create with DeepSeek. Keep pushing those boundaries and sharing your projects!

@FireworksAI_HQ Could this setup conceivably replace Cursor? 👀

@FireworksAI_HQ Thanks, always great information, appreciate you taking the time to share 🤜🤛