Grok Model Version Comparison
On May 1st, two data points landed in quick succession, and together they paint a portrait of a model that is better in some ways and noticeably worse in others. The first signal: Grok 4.3 shows higher overall intelligence scores compared to Grok 4.20, and it gets there at lower cost. That sounds like a clean win. The second signal complicates it immediately. On the Extended NYT Connections Benchmark — a test of language reasoning and categorical thinking — Grok 4.3 drops from 93.4 to 67.5 compared to Grok 4.20 0309. That is not a small regression. That is roughly a quarter of its performance on one specific measure, gone. Add to that a slightly elevated hallucination rate, and the picture that emerges is a model that has been optimised for efficiency and broad capability at some cost to precision.
If confirmed, here is what this means. xAI is making a trade-off that every major AI lab eventually faces: raw benchmark performance versus cost-per-output. Grok 4.3 appears to be the leaner, faster, cheaper option — but users who depend on precise language reasoning, the kind that powers editorial tools, categorisation tasks, or structured data work, may find it a step backwards. The hallucination uptick matters more than it sounds. In high-stakes applications, a slightly higher confabulation rate is not a minor inconvenience — it is a compounding liability. The strategic read here is that xAI may be targeting volume deployment and developer adoption over best-in-class accuracy, which tells you something about where they think the market is heading.
Watch for independent benchmark replications from researchers outside the Singularity ecosystem, and pay close attention to whether xAI addresses the NYT Connections regression directly — silence on that specific drop would itself be informative.
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1t0bdh3/grok_43_is_out_in_the_api/"> <img alt="Grok 4.3 is out in the API"
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1ura8v8/grok_45_has_an_advantage_on_cursorbench_an/"> <img alt=""Grok 4.5 has an advantage on
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1ur06sj/grok_45_is_live/"> <img alt="Grok 4.5 is live"
<a href="https://news.ycombinator.com/item?id=49789558">Comments</a>
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1ur6bie/grok45_on_par_with_gpt55xhigh_in_coding_at_half/"> <img alt="Grok-4.5 on par with gpt-5.5-xhigh in
<!-- SC_OFF --><div class="md"><p>I never thought there would be a model worse than the Gemini series, but here we are.</p> <p>I remember Elon tweeting that this model would beat
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1t1hjeo/grok_43_strong_in_finance_and_longcontext_with/"> <img alt="Grok 4.3: strong in finance and
<table> <tr><td> <a href="https://www.reddit.com/r/ChatGPT/comments/1vn100y/grok_46_finally_out_a_week_late_but_absolutely/"> <img alt="Grok 4.6 finally out, a week late but
NewsHive monitors these sources continuously. All signal titles above link to the original reporting.
Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →