{
  "schema_version": "1.1",
  "id": "archive:https://cursor.com/blog/grok-4-6",
  "slug": "grok-4-6-1m70nv0",
  "url": "https://feed7.dev/p/grok-4-6-1m70nv0",
  "title": "Introducing Grok 4.6",
  "why_included": "Grok 4.6 targets long-running coding and knowledge-work agents, with more self-testing and stronger visual first passes reported by Cursor. API pricing starts at $2 input and $6 output per million tokens.",
  "summary": "**Grok 4.6** is available in Cursor, Grok Build, the SpaceXAI API, and partner platforms. Cursor positions it for long-running coding and knowledge work, reporting more self-testing and stronger first passes on interactive and visual projects than Grok 4.5.",
  "practical_implication": "Builders should trial it on multi-step implementation, research, and UI prototyping, then compare task completion and verification behavior against their current model. Pricing starts at **$2/M input tokens** and **$6/M output tokens**.",
  "agent_context": "**Grok 4.6** is available in Cursor, Grok Build, the SpaceXAI API, and partner platforms. Cursor positions it for long-running coding and knowledge work, reporting more self-testing and stronger first passes on interactive and visual projects than Grok 4.5.\n\nBuilders should trial it on multi-step implementation, research, and UI prototyping, then compare task completion and verification behavior against their current model. Pricing starts at **$2/M input tokens** and **$6/M output tokens**.\n\nThe benchmark discussion supplies few raw scores; the main comparison says it matches GPT-5.6 Sol on a nine-benchmark composite. A fast variant costs **2x**, and Cursor’s qualitative project findings are not independent evaluations.",
  "source": {
    "name": "Cursor",
    "url": "https://cursor.com/blog/grok-4-6",
    "published_at": "2026-08-12T00:00:00.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Engineering Post",
  "layer": "model",
  "domains": [
    "coding",
    "research"
  ],
  "topics": [
    "reasoning",
    "model-selection",
    "coding-agents"
  ],
  "verification": {
    "status": "official_source",
    "label": "Official Source",
    "method": "source_feed",
    "verified_at": null
  },
  "uncertainty": [
    "The benchmark discussion supplies few raw scores; the main comparison says it matches GPT-5.6 Sol on a nine-benchmark composite. A fast variant costs **2x**, and Cursor’s qualitative project findings are not independent evaluations."
  ],
  "connected_context": {
    "meaning": "This advances Grok’s long-running coding and knowledge-work case from 4.5 with claimed improvements in self-testing and interactive or visual first passes. The GPT-5.6 Sol composite supplies a comparison target, but sparse raw scores and Cursor’s non-independent observations leave workload trials as the decision basis, especially when weighing standard versus 2x-priced fast service.",
    "corpus_size": 479,
    "generated_at": "2026-08-18T10:04:01.168Z",
    "connections": [
      {
        "title": "Introducing Grok 4.5",
        "source_name": "Cursor",
        "source_url": "https://cursor.com/blog/grok-4-5",
        "feed7_url": "https://feed7.dev/p/grok-4-5-1n0zgxx",
        "reason": "Grok 4.6 is presented as improving 4.5’s first-pass and self-testing behavior, while 4.5’s training-contamination issue remains a warning against relying on CursorBench comparisons alone."
      },
      {
        "title": "GPT 5.6 Sol, Luna, and Terra now available on AI Gateway",
        "source_name": "Vercel",
        "source_url": "https://vercel.com/changelog/gpt-5-6-now-available-on-ai-gateway",
        "feed7_url": "https://feed7.dev/p/gpt-5-6-now-available-on-ai-gateway-106pgsr",
        "reason": "GPT-5.6 Sol is the stated peer on the nine-benchmark composite and therefore the most direct candidate for representative task comparisons, despite the absence of detailed component scores."
      },
      {
        "title": "The Base Model Is Dead — Varun Singh, Arcee AI",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=xbPriQWXtWM",
        "feed7_url": "https://feed7.dev/p/the-base-model-is-dead-varun-singh-arcee-ai-02hts76",
        "reason": "The training-data argument reinforces evaluating Grok 4.6 on intended agent workloads rather than inferring capability from model branding or a composite benchmark."
      },
      {
        "title": "Introducing Claude Sonnet 5",
        "source_name": "Anthropic",
        "source_url": "https://www.anthropic.com/news/claude-sonnet-5",
        "feed7_url": "https://feed7.dev/p/claude-sonnet-5-1jlv18h",
        "reason": "Sonnet 5 provides another similarly priced coding-oriented option, but its tokenizer change means token counts and effective task costs must be measured rather than compared from headline per-token prices alone."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-08-12T00:00:00.000Z",
  "modified_at": "2026-08-12T00:00:00.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/grok-4-6-1m70nv0",
    "json": "https://feed7.dev/p/grok-4-6-1m70nv0.json",
    "markdown": "https://feed7.dev/p/grok-4-6-1m70nv0.md"
  }
}