Tested 7 d.Revised Apr. 26
Tool review · Detailed verdict
OpenAI
ChatGPT (GPT-5.5)
« Native omnimodal Swiss army knife. Back on top for agentic coding and long-context reasoning. »
Overall score
weighted average · 5 axes
§ Presentation video · ChatGPT (GPT-5.5)
YouTube · officiel
§ Product preview
Price / month
22 €
Pro plan
Cost / 1M tokens
14 €
output rate
Context
1000k
max tokens
GDPR
6.2/10
to verify
Hosting
🇺🇸 US
by default
§3.1Strength profile
Scale 0 to 10 · RadarOnAI internal method
§3.2+ Strong points
- 01Native omnimodal architecture (text, image, audio, video)
- 0282.7% Terminal-Bench 2.0
- 031M tokens, Codex built in
§3.3− Weak points
- 01Murky training data
- 02GDPR: US servers by default
Best for
Teams that want a single tool for everything.
Not for
Strict DPOs, missions covered by professional secrecy.
§04Benchmarks — ChatGPT (GPT-5.5)
Against the rest, cold.
BenchmarkScoreValueRank
GPQA Diamond
Scientific reasoning
80.1%
#2
SWE-bench Verified
Real-world bug fixing
68.2%
#3
MMLU-Pro
General knowledge
84.1%
#2
HumanEval+
Code generation
89.8%
#3
FR-Check 2026
Internal language test
84/100
#3
Overall score evolution · Q1 '25 → Q3 '26
Q1 '25Q2 '25Q3 '25Q4 '25Q1 '26Q2 '26Q3 '26
+26
pts in 7 quarters
§ Editor's view · try it?
If your long-form text matters more than your images, ChatGPT (GPT-5.5) remains the subscription that pays back its 22 € monthly fee fastest. We use it daily in the newsroom.
Affiliate link ↗ — RadarOnAI earns a commission if you subscribe, at no extra cost to you. Our verdicts never depend on commissions.
Essayer ChatGPT ↗
No commitment · cancel in one click