🇺🇸 🇨🇳 🇰🇷 Chatbot Accuracy by Model & Rates of Errors

Recent tests by the European Broadcasting Union found that artificial intelligence assistants misrepresented news content in 45% of evaluated cases across languages and regions, highlighting persistent concerns about accuracy as Al adoption grows.

Two concepts are measured here: (1) accuracy, or the rate at which responses were factually correct and (2) hallucination, or the proportion of responses a model gets wrong because it made them up:

Accuracy remains limited. Gemini 3 Preview (High) leads at 54%, followed by Claude Opus 4.5 at 43% and Grok 4 at 40%. Gemini 2.5 Pro reaches 37%, GPT-5.1 (High) 35%, and Claude 4.5 Sonnet 31%. Most other models fall into the twenties or teens, showing that higher accuracy does not automatically prevent frequent errors.

Hallucination rates vary widely. Claude 4.5 Haiku reports the lowest rate at 26%, followed by Claude 4.5 Sonnet at 48% and GPT-5.1 (High) at 51%. Claude Opus 4.5 reaches 58%...Grok 4 records 64%, Kimi K2 0905 69%, and Grok 4.1 Fast 72%. Kimi K2 Thinking reaches 74%, and Llama Nemotron Super 49B v1.5 76%.

DeepSeek models are among the least reliable. V3.2 Ex records 81%, R1 0528 83%, and EXAONE 4.032B 86%. Llama 4 Maverick posts 87.58%, while multiple Gemini variants exceed 87%. GLM-4.6 and gpt-oss-20B (High) top the chart above 93%.

This is why ChatGPT (OpenAI) and Claude (Anthropic) are doing well. The latter especially is making inroads into the commercial sector because of the high value placed on both of these two accuracy metrics.

It makes no difference what extra mile your model can pull off if it's inventing things when it should be uncertain. Grok 4 was the most accurate until fairly recently, but are companies going to switch to the probable new leader in accuracy, Gemini 3, when it hallucinates 88% of the time?

Related: AI Safety Rankings

🔥 I N T E L R U N N E R 🔥

🇺🇸 Share of U.S. respondents who have used the following Al tools in 2023, 2024, and 2025.

ChatGPT is the top answer in all three. Gemini has been rallying. Llama slides in obscurity as deserved. Claude & Mistral are solid products.

I distrust Microsoft more than I distrust Big Tech in general basis (quite a bit) and won't allow it on my devices.

⚡️ I N T E L R U N N E R ⚡️

🇭🇳 The prediction market for the presidential election in Honduras has shifted decidedly toward Salvador Nasralla with 53% odds.

What a wild race though. Look at the past month!

🔮 I N T E L R U N N E R 🔮

🇺🇸🇧🇷 The historic breakout in the Latin America 40 ETF continues. $ILF is up 49.74% YTD.

The top four holdings in $ILF are easily traded on American exchanges and they're all Brazilian:

🆎 Itaú Unibanco Holdings S.A. ($ITUB, 7.33%)
🆎 Vale S.A. ($VALE, 8.14%)
🆎 Petróleo Brasileiro S.A. (8.85% b/w $PBR & $PBRA)
🆎 Nu Holdings Ltd. ($NU, 11.78%)

📈 I N T E L R U N N E R 📈

🇺🇸 The Q4 2025 Cox Automotive Dealer Sentiment Index

CADSI shows the current market index falling -5 pts to 38 (from 43 in Q3), with the future outlook down -4 pts to 42, underscoring a broad deterioration in dealer sentiment heading into year-end.

🤩 I N T E L R U N N E R 🤩

🇺🇸 A third of the NASDAQ Composite stocks have triggered buy signals on the MACD.

We have not seen that since before & after Liberation Day.

📈 I N T E L R U N N E R 📈

🌍 Dividend yields are lower in America than almost everywhere else on Earth...

Exception: Argentina 🇦🇷

Highest Yields: Brazil 🇧🇷

📊 I N T E L R U N N E R 📊

🇺🇸 Gas Prices in America have slipped under $3 per gallon for the first time in 4 years.

It has certainly been a priority of the Trump Administration to attack oil prices.

Goldman Sachs expects the glut to continue into next year. I regard such predictions with suspicion given the source...

⛽️ INTELRUNNER ⛽️

🇺🇸 The AAII Asset Allocation to equities is the highest (71.2%) since the November 2021 peak.

Cash allocation is near 4-year lows (14.8%), and the allocation to bonds is slightly south of that, although not quite at the lowest lows...

📈 INTELRUNNER 📉

🇺🇸 Nvidia has been down 12-13% over the past month, but the S&P 500 and the Magnificent 7 have held up better. $SPX and $MAGS gained 14.3% and 11.75% respectively against $NVDA over the same period.

In other words, the market can survive at least one month without its leader...

More: A prescient post I made being cynical about Nvidia's valuation on the night of its all-time high.

📹 INTELRUNNER 📹

🇺🇸 The largest cumulative losses before young companies turned a profit compared with the cash burn for OpenAl and Anthropic based on media reports (in billions of dollars)...

You see why Sam Altman wants federally guaranteed loans? Let me state that differently:

Do you see why he wants the taxpayers to pick up his risk while he retains the reward? For his one-time "nonprofit" company?

🤩 I N T E L R U N N E R 🤩

🌍 Central banks are buying gold like there's about to be a world war or a financial reset—or something...

More on Gold
• Global Gold Reserves: the United States of America vs. the World [9/26]
• Gold's Best Year Since '79 (Plus A Review of Gold Coverage Here) [10/2]
• The Gold/Silver Ratio Drops [12/2]

🤩 I N T E L R U N N E R 🤩

🇺🇸 Ratio of the Leading Economic Index to the Coincident Economic Index

These are the two Conference Board indices tracking leading versus current indicators. Historically, this is the second worst ratio ever recorded—and counting.

The worst ever, narrowly, is the trough of the Great Financial Crisis, February through April of 2009. It has now surpassed the 1981-82 recession.

As I noted in August, these levels tend to mark the terminus of recessions—you know, the recession we're "not in" because "my 401K is fine."

📉 I N T E L R U N N E R 📉

🇺🇸 Current Locations of U.S. Carrier Strike Groups (as of December 2nd)

Carrier strike group (CSG) is an operational formation of the United States Navy. It is centered on an aircraft carrier and a carrier air wing of 65 to 70 aircrafts. It’s composed of roughly 7,500 personnel, an aircraft carrier, at least one cruiser, a destroyer squadron of at least two destroyers and/or frigates. A carrier strike group also, on occasion, includes submarines, attached logistics ships and a supply ship. Carrier strike groups comprise a principal element of U.S. power projection over the world’s oceans.

Well yeah, okay, sure, depending on whom you're projecting power at.

They're not all that useful against geopolitical rivals anymore. If you're Somali pirate, then yes, you're done for. But those aren't the only types of people D.C. is picking fights with these days.

When it comes to those sorts of nations, a much bigger submarine fleet and a whole lot of drone production will be a lot more valuable. Unfortunately, that's not how budgetary decisions are made in D.C.

It has far more to do with Senator So & So wanting aircraft carrier production to continue in his state. And of course the complex is always fighting amongst itself over the trillion dollars with little regard for anything but Wall Street's latest analysis...

💫 I N T E L R U N N E R 💫

🇪🇺 Can EURUSD sustain this two-week trend?

Decent odds. The pound looks ready for a bounce itself.

But not advice, don't forget it.

🔎 I N T E L R U N N E R 🔎

🇺🇸 A visualization of the morning session in yesterday's market. Each set of bubbles represents a sector.

What this shows more than anything is how comically unbalanced the S&P 500 ($SPX) is. That's okay though.

It's not like energy, utilities, or basic materials are significant parts of the economy or anything...

🟢 I N T E L R U N N E R 🟢

🇺🇸 Natural gas has run quite a ways now. NG1! topped out at 108.3% in 102 days before closing up 100.4%.

I've got a gut instinct we're due for a pullback here.

Absolutely not financial advice...

✅ I N T E L R U N N E R ✅

🇺🇸 Change in Forward P/E of S&P 500 Industry Groups (Since 10/28/2025 High)

Semiconductors and software have been getting killed...not so hot for auto either.

Media & entertainment, pharma & biotech, and retail & food were the only industry groups above 1X.

📊 I N T E L R U N N E R 📊

🌍 The Baltic Dry Index tracks the cost of shipping raw materials like iron ore, coal, and grain across various sea routes.

The BDI is often used as an indicator of future economic activity, understandably, because it signals an intent to produce.

However, it's also a warning sign of potential inflationary pressures. The BDI is up 165% YTD and it hit a local high 2,845 yesterday—the highest in 2 years.

📈 I N T E L R U N N E R 📈