Grok 4.7 Coding Tests: Strong Tasks, Uneven Everyday Results

AICodeKing11m 47s
A flat balance illustration weighs a check mark against stacked work blocks beside the blue-and-white headline “BENCHMARKS VS REAL WORK” on black.
0 comments · 0 votesOpen discussionClose discussion
Sign in to join the discussion

    The AICodeKing presenter contrasts reported benchmark gains with the cost and time needed to complete an agent taskCost per completed task measures the total AI, tool, infrastructure, retry, and repair expense for each verified useful outcome.. He distinguishes per-token pricingToken pricing is the rate an AI provider charges for processing input tokens, generating output tokens, or reading cached tokens. from overall usage and notes that reasoning settings and harness choicesAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction. affect comparisons. The video presents those measurements as evaluation results, not a promise about every user's bill.

    In eight personally scored tasksEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria., Grok 4.7 performs well on a contact-lens case, a folding table and an end-to-end model-training exercise, but struggles with an SVG illustration and a detailed wristwatch. The presenter describes these scores as his own judgments and acknowledges that stronger engineering results can coexist with weak visual consistency.

    The AICodeKing presenter reports missed requirements and repetitive loops in everyday frontend work. He suggests trying the model on narrowly defined assignmentsAgent task scoping defines the objective, boundaries, inputs, permissions, deliverables, and completion evidence for an AI agent assignment. under a stronger planner and reviewer, while checking whether extra tokens and rework erase any savings. Interface endorsements and channel-support promotions are omitted.

    Original YouTube thumbnailWatch on YouTube

    Video topic coverage

    • Agents & workflows85
    • Coding & tools100
    • Models & learning85
    • Infrastructure & compute0
    • Research & evaluation95
    • Safety & security0
    • Business & productivity70
    • Creative & media45
    • Robotics & perception0
    • Society & policy0

    Share this video

    A flat balance illustration weighs a check mark against stacked work blocks beside the blue-and-white headline “BENCHMARKS VS REAL WORK” on black. With WWW.ARTIFICIAL-INTELLIGENCE.VIDEO, 22 September 2026 and duration 11m 47s.

    AICodeKing finds useful engineering capability in Grok 4.7 but argues that uneven visual output, repeated steering and higher completed-task costs make it a difficult default.