Google is no longer competing with OpenAI and Anthropic to reach the same finish line. Its competitors want AI that improves itself. Google wants AI that understands the real world.
It’s easy to miss this dichotomy, because Google still ships models and still makes money. But her latest release ranked 10th on one independent rating.
What Google said and didn’t say about the AI race
Google released Gemini 3.6 Flash on July 21. It was supply speed and cost, not raw power.
Follow us on X To get the latest news as it happens
The model produces 17% fewer codes than its predecessor, the Google version Blog He said. Codes are small pieces of text written by artificial intelligence. Fewer codes means a cheaper answer.
Authority is another matter. One published reading of the Synthetic Analysis Index placed the model 10th. All other major laboratories ranked higher than him.
Google doesn’t stand still. It has begun its largest training run to date, for Gemini 4. The larger model, Gemini 3.5 Pro, is still in testing with partners.
Sundar Pichai pointed to a different award. He’s hooked up Roadmap for personal agents Instead of winning the leaderboard.
Investors are less relaxed. Alphabet shares fell 6% in June after two big… Researchers left for competitors.
Inside DeepMind’s bet on global models
The universal model is artificial intelligence that learns how physical objects behave. Gravity, motion, cause and effect. It predicts what will happen next in the room, not the next word in the sentence.
DeepMind’s own site offers the bet. It positions Genie 3 and Gemini Robotics under the rubric of universal models and embodied artificial intelligence, that is, programs that control machines.
In May, the lab expanded Project Genie to include Street View. It also released SIMA 2, an agent that learns by playing within 3D virtual worlds.
OpenAI and Anthropic are aiming somewhere else entirely. They want iterative self-improvement, shortened to Relative Strength Index (RSI).
In plain terms, this AI is smart enough to build the next best AI. Then he builds the next one.
Writer Alberto Romero said on Tuesday that Google had intentionally left this race.
“Hasabis is betting on something else: universal models. Models that can understand and simulate the real world, not just predict the next code,” Alberto Romero books In a recent analysis.
Google didn’t say anything like that. Demis Hassabis, who runs Google DeepMind, has never dismissed the Relative Strength Index (RSI) in public.
Why would a rival founder say that DeepMind is the odd one out?
The sharpest outside reading came from a competitor months ago.
Jack Clarke co-founded Anthropic. On May 4 he published article Where is artificial intelligence headed?
He gave a 60% chance that AI could do its own research by the end of 2028. He put 2027 at 30%.
Clark then asked which laboratories were pursuing this goal. DeepMind “appears to be the most cautious of the Big Three,” he wrote. Caution means caution.
His evidence came from DeepMind itself. He pointed to the year 2025 paper About AI Safety, co-written by co-founder Shane Legg.
Anthropic is much bolder. It was reported in A Recursive self-improvement A study suggests that Cloud wrote more than 80% of the code it ships by May 2026.
Before February 2025, this share was close to zero.
The company has documented AI builds better AI. In one speed test, its models achieved 52x gains in April, versus 2.9x the previous year.
A skilled engineer needs four to eight hours to achieve four times the gain on the same task.
Why might Google not withdraw from the AI race at all?
There are two facts that contradict the whole idea:
- Google leads the test that comes closest to measuring search skills in the field of artificial intelligence.
MLE-Bench asks the model to build machine learning systems on its own. Gemini 3 model received a rating of 64.4% in February. That was the best result at that time. Google put Flash 3.6 up by 63.9% in July.
Labs that leave a field rarely top the scoreboard.
- Google is not absent from the market.
Pichai told investors that the Gemini app has 950 million monthly users. Google skipped it NVIDIA Open AI Alliance This month. So did OpenAI and Anthropy.
Can Google afford to wait?
The issue of patience is simple. Search ads pay for everything else, so DeepMind can take its time.
Alphabet’s own profile shows that the pillow has become thinner.
Revenue reached $119.8 billion in the June quarterAn increase of 24%, according to results Submitted on July 22. The research alone brought in $63.3 billion.
Then comes spending. Alphabet has pumped $44.9 billion into data centers and equipment in three months. This is almost double what it was a year ago.
The result was negative free cash flow of $5.86 billion. Free cash flow is what’s left over after paying construction bills.
This number was positive $10.1 billion in March. In December it was positive $24.6 billion.
Alphabet covered the gap by selling $49.6 billion worth of new shares in June. It borrowed another $20.3 billion.
Long-term debt doubled in six months, from $46.5 billion to $98.2 billion.
A line in accounts covering corporate AI research lost $5.79 billion, up from $3.37 billion. Patience now has a price.
What to watch for in the next 30 days
- Whether the Gemini 3.5 Pro charges and how it’s rated
- Whether DeepMind displays global model results associated with Gemini 4
- Whether Alphabet’s cash flow turned positive again in September remains to be seen
- Whether Hassabis answers the self-improvement question directly
Gemini 4 is the real test. If global models succeed where programming agents fail, the slow pace will seem smart rather than scary.
The next earnings report will show how long Alphabet can keep paying to find out.
this post Why might Google quietly leave the AI race to OpenAI and Anthropic? appeared first on BeInCrypto.
