A Hands-On Test of Google’s Newest AI

Simply three months after the discharge of its cutting-edge mannequin, Gemini 3 Professional, Google DeepMind has launched its newest model, Gemini 3.1 Professional.

Considerably upgraded by way of performance and security, the Gemini 3.1 Professional mannequin strives to be accessible and operational for everybody. No matter choice, platform, or buying energy, this mannequin affords many advantages to all customers.

I can be testing the options of Gemini 3.1 Professional and detailing its key options. Every thing it’s good to find out about this new mannequin, from find out how to entry Gemini 3.1 Professional to benchmarks, is roofed on this article.

Gemini 3.1 Professional: What’s new?

Gemini 3.1 Professional is the most recent member of the Gemini mannequin household. As all the time, this mannequin has an unimaginable quantity of options and enhancements over the previous. Probably the most notable ones are:

1M Context Window: Maintains an industry-leading 1M token enter capability, permitting you to course of over 1,500 pages of textual content or a complete code repository with a single immediate. Superior inference efficiency: Delivers greater than twice the inference efficiency of Gemini 3 Professional, scoring 77.1% on the ARC-AGI-2 benchmark. Enhanced agent reliability: Particularly optimized for autonomous workflows, together with precision software orchestration and devoted API endpoints (gemini-3.1-pro-preview-customtools) for bash execution. Value: The associated fee/tokens of the most recent mannequin are the identical because the earlier mannequin. Those that are conversant in the Professional model can improve free of charge.

Superior Vibe Coding: This mannequin handles visible coding very nicely. Generate animated SVGs in your web site utilizing simply code, with crisp scaling and small file sizes. Hallucinations: Gemini 3.1 Professional tackles the issue of hallucinations head-on by lowering the hallucination price from 88% to 50% throughout AA-Omniscience. Data and Hallucination Benchmarks Granular Considering: This mannequin provides extra granularity to the pondering choices offered by its predecessor. Customers can now select between excessive, medium, and low pondering parameters.

Palms-on: Have enjoyable

It does not matter how a lot it will get talked about all around the world if it really does not carry out nicely. To correctly consider Gemini 3.1 Professional, we examined it throughout three classes:

Complicated inference Code technology and debugging Lengthy context synthesis

Process 1: Multi-step logical reasoning

This assessments: thought chain reasoning, constraint processing, and hallucination tolerance.

immediate:

“You might be given the next state of affairs.

5 analysts (A, B, C, D, and E) are assigned to 3 initiatives: Alpha, Beta, and Gamma.

rule:

1. Every mission requires at the least one analyst.
2. A can’t work along with C.
3. B have to be assigned to the identical mission as D.
4. E can’t be on Alpha.
5. Not more than three analysts could also be assigned to any mission.

Query: Please checklist all legitimate project mixtures. Be clear about your causes and be sure to’re not breaking any guidelines.

“

response:

Gemini 3.1 Professional dealt with the constraining logic with out falling into the inconsistencies that journey most fashions up. Consistency and readability in enumerating legitimate mixtures demonstrated critical reasoning depth.

Process 2: Code technology and refactoring

This assessments: Structured Considering, Edge Case Recognition, Code Correctness, and Manufacturing Actuality.

immediate:

“Given the next Python operate.

def process_orders(orders): complete = 0 for order so as: if order[“status”] == “Full”: Whole += Order[“amount”]
Return complete

Refactor this operate as follows:

1. Securely deal with your misplaced keys.
2. Adverse quantities are ignored.
3. Log skipped invalid entries.
4. Make it production-ready utilizing sort hints and docstrings.
5. Embody unit assessments for edge instances.

Returns solely clear, well-structured Python code.

“

response:

Import unit check from import log enter import Record, Dict, Any logger =logging.getLogger(__name__) def process_orders(orders:Record[Dict[str, Any]]) -> float: “”” Course of the checklist of order dictionaries and calculate the entire quantity of accomplished orders. “”” complete: float = 0.0 if not isinstance(orders, checklist): logger.error(“‘orders’ needs to be a listing.”) return complete forindex, order in enumerate(orders): if not isinstance(order, dict): logger.warning(“Skip index %d: Not a Dictionary.”,index) order.get(“standing”) != “accomplished”: Proceed if “quantity” is out of order: logger.warning(“Skipping Index %d: Lacking ‘quantity’.”,index) proceed quantity = order[“amount”]

If not isinstance(quantity, (int, float)) or isinstance(quantity, bool): logger.warning(“Skip index %d: ‘quantity’ have to be numeric.”, Index) If quantity < 0, proceed: logger.warning("Skip index %d: Adverse quantity (%.2f).", Index, quantity) Proceed complete += float(quantity) return complete class TestProcessOrders(unittest.TestCase): def setUp(self):logging.disable(logging.WARNING) def TeaDown(self):logging.disable(logging.NOTSET) def test_valid_completed_orders(self):orders = [ {"status": "completed", "amount": 100}, {"status": "completed", "amount": 50.5} ] self.assertEqual(process_orders(orders), 150.5) def test_ignore_pending_orders(self):orders = [ {"status": "pending", "amount": 100}, {"status": "completed", "amount": 50} ] self.assertEqual(process_orders(orders), 50.0) def test_missing_keys_skipped(self):orders = [ {"amount": 100}, {"status": "completed"}, {"status": "completed", "amount": 20} ] self.assertEqual(process_orders(orders), 20.0) def test_negative_amounts_ignored(self):orders =[{"standing":"accomplished""quantity":-10}{"standing":"accomplished""quantity":3

I felt that the refactored code was extra geared in the direction of a manufacturing surroundings than a toy degree. Anticipated edge instances, enforced sort security, and inbuilt significant assessments. That is the form of output that truly respects real-world improvement requirements.

Process 3: Lengthy context evaluation synthesis

This assessments: data compression, structured summarization, and reasoning throughout contexts.

immediate:

“The next is a complete enterprise report.

Firm Identify: NovaGrid AI

2022 income: $12 million
2023 income: $28 million
2024 income: $46 million

Buyer churn elevated from 4% to 11% in 2024.
R&D spending elevated by 70% in 2024.
Working margin decreased from 18% to 9%.
Enterprise prospects elevated by 40%.
SMB prospects had been down 22%.
The price of cloud infrastructure has doubled.

process:

1. Diagnose the probably root reason behind margin decline.
2. Determine strategic dangers.
3. Suggest three actions based mostly on the information.
4. Current your reply in a structured government memo format.

“

response:

This tied monetary indicators, operational adjustments, and strategic dangers right into a constant administration narrative. The flexibility to diagnose margin stress whereas balancing development indicators represents sturdy enterprise reasoning. It wasn’t a normal abstract, it was one thing a sensible strategic advisor would draft.

Word: I didn’t use the usual “Create Dashboard” process as most fashionable fashions similar to Sonnet 4.6, Kimi Okay 2.5 and so on. permit you to simply create a dashboard. So for a mannequin this succesful, it should not be too troublesome.

How do I entry Gemini 3.1 Professional?

In contrast to earlier Professional fashions, Gemini 3.1 Professional is freely accessible to all customers on their platform of selection.

Now that we have determined to make use of Gemini 3.1 Professional, let’s check out find out how to entry the fashions.

Gemini Internet UI: Free and Gemini Superior customers now have 3.1 Professional obtainable within the mannequin part choices.

API: Accessible by way of Google AI Studio for builders (mannequin/Gemini-3.1-pro).

Cloud platforms: Deployed on NotebookLM, Vertex AI on Google Cloud, and Microsoft Foundry.

benchmark

Benchmarks are helpful to quantify how good this mannequin is.

There’s quite a bit to decipher right here. However essentially the most superb development of all is unquestionably in summary reasoning puzzles.

Let’s put issues into perspective. Gemini 3 Professional launched with an ARC-AGI-2 rating of 31.1%. This was thought of the perfect on the time and a breakthrough within the LLM customary. Quick ahead simply three months, and its rating has surpassed its successor by an element of two.

Which means AI fashions are bettering quickly.

If you happen to’re undecided what these benchmarks check, learn our AI benchmarks article.

Backside line: highly effective and accessible

The Gemini 3.1 Professional proves to be greater than only a fancy multimodal mannequin. Reveal production-relevant real-world performance throughout inference, code, and analytical synthesis. It isn’t good and nonetheless requires structured prompts and human supervision. However as a frontier mannequin constructed into Google’s ecosystem, it is sturdy, aggressive, and value critical consideration.

FAQ