Grok 3.5 Adds Real-Time Web Search, Native Image Gen, and 256K Context
xAI has released Grok 3.5 with real-time web search grounding, native image generation via the Aurora model, and a 256K context window. The update is available to X Premium+ subscribers and through the xAI API.
Original sourcexAI has shipped Grok 3.5, a meaningful capability expansion to its flagship model that adds three distinct features: real-time web grounding for up-to-date answers, native image generation powered by xAI's Aurora image model, and a doubled context window now sitting at 256K tokens. The update positions Grok more directly against multimodal, search-augmented models like GPT-4o and Gemini 1.5 Pro.
The real-time web grounding is the headline addition — Grok can now pull live information from the web during inference, reducing reliance on static training data for time-sensitive queries. Combined with the 256K context window, this makes the model considerably more useful for tasks that require synthesizing large volumes of current information. Aurora-powered image generation is integrated natively, meaning users and API consumers don't need a separate pipeline to go from text to image.
Access is split across two tiers: X Premium+ subscribers get the consumer-facing Grok experience on the platform, while developers can access the model via the xAI API. Pricing for the API tier has not been prominently detailed in the announcement. The Aurora image generation capability was previously available as a standalone feature on X and is now unified under the Grok 3.5 release.
The release continues xAI's strategy of building Grok as a tightly integrated feature of the X platform while simultaneously pursuing API-first developer adoption. Whether the model's capabilities are competitive with frontier alternatives remains to be tested independently — xAI has not published third-party benchmark results alongside this release.
Panel Takes
The Builder
Developer Perspective
“The primitive here is a single API endpoint that gives you grounded search, image generation, and 256K context — which is genuinely useful if the implementation isn't a mess of separate endpoints and auth flows. What I need to know before I care: is web grounding a parameter flag or a separate model variant, and does the image generation return URLs or base64 blobs with no lifecycle guarantees? xAI still has a documentation problem — the API surface is real but the developer experience has historically felt like an afterthought relative to the X consumer product, and I'll believe this is first-class API design when I see the actual request schema, not a blog post.”
The Skeptic
Reality Check
“This is xAI playing catch-up to a checklist that OpenAI and Google already shipped — web search grounding, image generation, big context window. The competitive moat here isn't the features, it's the X platform distribution, and that's a ceiling as much as a floor because X's user base is not the same as the developer or enterprise buyer you need to build a durable AI business. What kills this in 12 months isn't a better competitor — it's that Grok's primary value prop is real-time X data, and if that data advantage doesn't show up meaningfully in benchmark tasks that buyers actually care about, this is just a more expensive commodity model with a bird logo.”
The Creator
Content & Design
“Aurora's image output has been available on X for a while, and the honest description is that it produces competent, slightly over-saturated generations that sit somewhere between Midjourney v5 and DALL-E 3 in terms of aesthetic default — fine for quick social content, not something you'd ship in a brand context without significant prompting discipline. Bundling it with the language model is genuinely useful for content workflows where you want research and visuals in one pass, but the editing surface is still shallow — there's no real iteration loop, just regenerate-and-hope. Until I see a public output gallery or can test refinement workflows, the image gen is a checkbox, not a craft tool.”
The Founder
Business & Market
“The distribution strategy here is actually coherent in a way most AI companies can't replicate — you have a built-in consumer funnel through X Premium+ that subsidizes model development while the API tier chases developer revenue, and Elon's real-time X data is a genuine proprietary input that OpenAI can't just replicate. The problem is pricing opacity: if xAI buries the API cost behind 'contact for enterprise' and doesn't publish transparent token pricing, they're going to lose the developer adoption flywheel to Anthropic and Google who have made their pricing tables boring and legible. The moat is the data and the distribution; the risk is that the product team is building for X the platform and the API is an afterthought.”