bugrasa (@bugrasa) header
bugrasa (@bugrasa) profile picture

Founder of @evaonlineai. One prompt, four AI models, side by side. Sharing what I learn building it in public.

Joined August 2010
Correct JSON wasn't enough to pick a CSV converter. All three models got our Nov 16 EVA sample right. Only Opus named Python's csv module as an alternative. I want that warning before reusing the code. Next test: a newline inside a quoted field.
10
GPT-6.1 Sol left the release date unapproved in a summary of fictional project notes. The fix was due Thursday. I want that separation kept. Before forwarding an AI summary, check which task each date actually belongs to.
7
A task list can look finished while a deadline is still missing. In this week's meeting-note test, GPT-6.1 Sol kept Leo's date unspecified. I want that gap visible before the list reaches a tracker. Check each date against the original notes.
1
13
Most surprising test this week: a 15-word hook prompt. Two of the four models made up a lost deal with a dollar amount ($20K and $10k). The prompt had no story in it. The test showed why short copy still needs a check.
4
In this test, four AI models agreed on the facts. They differed in tone and in what they added. One line, three readers. Claude and Gemini added a retention angle nobody asked for. Gemini also went with "supercharged". GPT-6 Astra kept each line to one sentence. Compare them before you trust one: evaonline.ai
43
The most useful result from this week's tests wasn't a wrong answer. All three models got the numbers right. One signed an email with a name nobody gave it. Another called a launch "on track" when the notes only said the date hadn't moved. Correct math is the easy part to check. Additions are the part to read twice.
7
I trust AI outputs more when I compare reasoning, not just answers. The winner is not always the model with the cleanest final response. I want to see which one frames the problem better. Which one misses constraints. Which one exposes a path I would not have considered. Side-by-side comparison is a work habit, not a leaderboard ritual. Compare reasoning structure when the decision matters. evaonline.ai
12
Running the same prompt through three models is not a workflow yet. Try it with something concrete, like support-ticket triage. One model may classify faster. One may explain the edge case better. One may hallucinate a policy that does not exist. The useful part is not staring at three answers. It is deciding which result changes the next routing rule, then saving that decision somewhere the next run can use it. Compare outputs with a decision rule, not just curiosity.
1
I got tired of choosing AI models based on vibes, so I ran the same 5 prompts through GPT-5.5, Claude Sonnet 4.6, and Gemini 3.1 Pro side-by-side. Same prompt. Same context. No follow-up corrections. The tests: 1. code debugging 2. long-form summarization 3. creative writing 4. factual / recent-info Q&A 5. explaining a complex concept simply Here’s what I found. For code debugging, GPT-5.5 was the strongest. All three caught the obvious JavaScript bug (`=` instead of `===`), but GPT caught more edge cases: bad coupon values, NaN, negative totals, currency precision, and whether coupon values were percentages or fixed amounts. Winner: GPT-5.5 > Claude Sonnet 4.6 > Gemini 3.1 Pro For long-form summarization, Claude Sonnet 4.6 won. I used an academic paper about entrepreneurial ecosystems and asked for a founder-focused summary. GPT-5.5 was more faithful to the paper, but Claude gave the more useful “busy founder” version. Less academic, more actionable. Winner: Claude Sonnet 4.6 > GPT-5.5 > Gemini 3.1 Pro Creative writing was the most subjective one. Prompt was a short story about a customer support agent realizing their AI tool knows more about their life than it should. GPT-5.5 had the best restraint. It stayed quiet and unsettling without turning into obvious horror. Claude was clean and readable. Gemini had some strong creepy moments, but pushed harder than I wanted. Winner: GPT-5.5 > Claude Sonnet 4.6 > Gemini 3.1 Pro For factual / recent-info Q&A, Claude won. This was interesting because the best answer was not the most detailed one. Claude was the most honest about uncertainty and said what it would verify before making a recommendation. GPT-5.5 was more useful, but made more assumptions. Gemini flagged uncertainty, then still gave a confident answer based on older equivalents. Winner: Claude Sonnet 4.6 > GPT-5.5 > Gemini 3.1 Pro For explaining vector databases to a non-technical founder, GPT-5.5 won. It explained the business value clearly, gave a realistic SaaS support example, clarified that vector DBs don’t “think” by themselves, and ended with when NOT to use one. Winner: GPT-5.5 > Gemini 3.1 Pro > Claude Sonnet 4.6 My takeaway: There isn’t one “best” model. GPT-5.5 is the best default when I want something complete and shippable. Claude Sonnet 4.6 is better when judgment, honesty, or readability matters. Gemini 3.1 Pro is good for simpler first-pass explanations, but felt less nuanced in these tests. The real pain is not model quality anymore. It’s workflow. Opening 3 tabs, pasting the same context, waiting, comparing manually, and remembering which model did better for which task is annoying. That’s why I’ve been using EVA Multi Chat lately: same prompt, multiple models, side-by-side outputs. Much easier to see the tradeoffs. Do you mostly stick to one model, or do you rotate depending on the task?
1
123
I've been running the same prompt through GPT, Claude, and Gemini every day for months while building EVA. What I kept seeing: → Claude follows complex instructions more reliably → GPT is more consistent with structured output → Gemini wins when you need recent information None of them wins everything. That's exactly why I built a workspace that runs all three at once. evaonline.ai
1
36
Replying to @tdinh_me
Building @evaonlineai - a unified Al workspace where people can use multiple top models (like Gemini, Claude, GPT) in one place.
1
64
Replying to @foundrceo
Building @evaonlineai - a unified AI workspace where people can use multiple top models (like Gemini, Claude, GPT) in one place. Compare outputs, avoid juggling subscriptions, use AI like an OS.
1
1
33
Replying to @heyblake
Currently building @evaonlineai One workspace to compare AI models side-by-side and pick the output that actually works. Early access: evaonline.ai
8
Most AI debates are pointless. “GPT is better” “No, Claude is smarter” “Gemini is underrated” None of this matters without seeing real outputs side-by-side. #AI #LLM #AItools
13