Is Grok-4.0 really the undisputed champion of instruction-following as claimed?

1. IMAGE DESCRIPTION
This is a screenshot of an X (formerly Twitter) post by @XFreeze, reposted by Elon Musk. The post claims: "Grok-4.0 just claimed #1 on IFBench (Artificial Analysis) – the gold standard for instruction following. 81% score. Outranking every other model." It explains that this means Grok follows instructions precisely without approximating or going off-script. It praises xAI for building the most reliable AI. A bar chart titled "Grok-4.0 Beta-2 ranks #1 in IFBench (Instruction Following)" shows a tall bar at 81% labeled for Grok-4.0 Beta-2 (in blue), higher than other models (e.g., bars at ~79%, 77%, 76%, down to ~43%), with labels like o4-mini, Claude variants, GPT, Gemini, Zephyr, etc., at the bottom.[1]

2. CLAIM
Grok-4.0 Beta-2 ranks #1 on IFBench (Artificial Analysis' instruction-following benchmark) with an 81% score, outranking all other models, as shown in the chart.[1]

3. VERDICT
MISLEADING

4. EVIDENCE
IFBench is a real benchmark by Artificial Analysis evaluating precise instruction-following on 58 diverse tasks.[1] The current leaderboard (as of 2026-04-12) tops with Grok 4.20 0309 (Reasoning) at 82.9%, Grok 4.20 0309 v2 (Reasoning) at 81.2%, and Nova 2.0 Pro Preview (low) at 79.6%.[1] No "Grok-4.0 Beta-2" appears on the leaderboard or in searches. Recent posts note Grok 4.20 Beta topping at 82.9%.[2] Grok 4 (base) was released in July 2025 but no IFBench score of exactly 81% for a "Beta-2" variant verified.[3] No matching X post by @XFreeze found.[1]

5. MANIPULATION CHECK
Possible editing: Model name "Grok-4.0 Beta-2" and exact 81% score do not match current leaderboard (closest is 81.2% for Grok 4.20 0309 v2). No bar chart on the IFBench page matches the image's layout or labels (e.g., no "Grok-4.0 Beta-2" bar). Chart may be outdated, cropped from a prior version, or labels altered—other bars align roughly with past scores, but no direct source confirms this graph.[1]

6. CREDIBILITY
5/10 (Grok models lead IFBench, but specific model/score/graph unverified; promotional hype.)

7. CONFIDENCE
9/10 (Direct leaderboard access and multiple searches confirm mismatch.)

8. BIAS
RIGHT (Pro-xAI/Elon Musk promotion.)

9. CATEGORY
Technology & AI