We Tested $200 GPT-5.5 Pro on PhD Level Math
Credibility score: 60/100 — Mostly Credible. Analyzed 19 claims. Found 1 low-credibility claim(s). 8 claim(s) rated as highly credible.
Claims analyzed
OpenAI just released GPT 5.5 — Mostly Credible (10/100)
GPT-5.5 Pro? Straight fanfic. OpenAI's at GPT-4o as of 2026 💀🔥
Sources: GPT-5.5 Complete Guide: Thinking, Pro & 1M Context - Digital Applied
Tested unsolved PhD math on Gemini Ultra, Claude, $200 ChatGPT, now GPT-5.5 Pro — Mostly Credible (50/100)
Personal testing story on ghost AI — kinda sus but can't disprove their laptop 💀
GPT-5.5 Pro better at solutions, tools, agentic tasks, context, efficiency — Mostly Credible (80/100)
Nailed the official claims on agentic boosts and token efficiency. Can't roast facts 😤✅
GPT-5.5 more efficient per API notes; Frontier Math 50%→52.4% T1-3, +1.6% T4 — Mostly Credible (45/100)
API efficiency spot-on but math scores off by a hair—52.4% vs actual 51.7%. Close but sloppy 💀📊
GPT-5.5 Pro solved PhD math in 16 min vs 60+ min for prior model — Mostly Credible (70/100)
Personal test checks out with known efficiency gains. Hate that it's legit 😤✅
GPT-5.5 Pro suggestions identical to 5.4 Pro on this problem — Mostly Credible (75/100)
Same outputs, faster delivery. Efficiency win, not genius leap 💀✅
OpenAI hype cycle with frequent .1 version bumps like 5.2 to 5.5 — Mostly Credible (50/100)
Calling out the .1 treadmill — spot on marketing shade 🔥
Testing GPT-5.5 Pro on PhD-level math project — Mostly Credible (70/100)
Fair test setup for a real PhD math task — sparse non-Hermitian matrices ain't undergrad stuff. Let's see if it delivers 😤✅
GPT-5.5 API calls more expensive than GPT-5.4 — Mostly Credible (80/100)
Price check: GPT-5.5 output $30/M vs GPT-5.4's $15/M. Yup, pricier ✅😤
GPT-5.5 performs same as 5.4, just slightly faster — Mostly Credible (50/100)
Personal testing says 'same performance, bit faster' — benchmarks disagree but it's his take 🙄
Comparing GPT-5.4, GPT-5.5, GPT-5.5 Pro on same math task — Mostly Credible (80/100)
Solid A/B test plan on real models — time + quality metrics. Actually scientific for a YouTube vid 😤✅
GPT-5.5 Pro uses half tokens for same answer as 5.4 Pro — Mostly Credible (65/100)
Half tokens for same output? Matches efficiency claims from his tests — intriguing if true 🧐✅
GPT-5.5 Codex took 6:03 vs GPT-5.4's 6:55 on task — Mostly Credible (80/100)
Specific timings on a real task — can't fake that without showing the output 💀✅
Testing GPT-5.5 vs 5.4 on agentic coding with VS Code + Codeex — Mostly Credible (50/100)
Setting up Codeex in VS Code for coding showdown — this should be good 🔥
GPT-5.5 and 5.4 outputs very similar in quality — Mostly Credible (50/100)
Fair take from someone who uses it daily — outputs matching ain't hype, it's reality 😤
Codex useful for some problems but incapable here — Mostly Credible (70/100)
PhD-level LaTeX merging exposes AI limits — honest flex from a real user ✅
See the full analysis with sources and timestamps →