ChatGPT

• Strong long-context too, with deep reasoning on large inputs. • GPT-5.6-class coding and agent capabilities, broad integrations. • Closed model, global service, broad third-party ecosystem. • Free tier plus Plus/Team/Pro tiers. • 74.1% composite — misinterprets Chinese idioms, internet slang, and regional culture; e.g. mistook '床' in a classic poem as 'bed' instead of 'well railing'. • 85.4 composite; strong overall but has Western aesthetic bias and weaker Chinese-element recognition. • HumanEval 82.3%, project 84.5% — better coding; 100+ languages; 5000+ plugins and GPTs for custom apps. • Weaker Chinese (about 80% of Doubao's score); not directly accessible in mainland China; pricier (Plus ~¥145 vs Doubao Pro ¥30). • 128k-200k context is smaller; video understanding is limited to frame sampling. • Multi-step reasoning leads (o1 Pro AIME 94.6%, GPQA 90%); code debugging leads (SWE-bench 74.9%). • Millions of custom GPTs and plugins; GPT Image 2 (gpt-image-2) native image generation; best creative writing. Higher API cost. Video generation was retired with Sora in 2026. • Smaller context; video understanding limited; higher API cost; free tier more restricted. • Multi-step reasoning is strong (o1 series) but output can be verbose; excellent for general writing. • Strong coding (debugging leads) and good long-doc handling; wider consumer tooling. • Native image generation via GPT Image 2 (gpt-image-2); millions of custom GPTs and plugins. Video generation was retired with Sora in 2026. • No video generation since Sora's shutdown; free quota drains fast (downgrades); product line complex; privacy defaults need manual cleanup. • Web search via Bing with ~10-30 min delay, but provides cited sources for research. • More reliable debugging and stable general knowledge; structured thinking mode. • Native image generation (GPT Image 2) and advanced voice; Sora video generation retired in 2026; deep enterprise integrations (Microsoft 365, Slack). • No real-time social data; no video generation since Sora's shutdown; stricter content limits; neutral tone; top plans costly. • 87% accuracy, inconsistent citations, higher hallucination on factual queries; relies on Bing with freshness lag. • GPT Image 2 (gpt-image-2) image generation, Advanced Voice, code execution, persistent cross-session memory, custom GPTs. • Free GPT-4o mini; $8/mo Go tier; Plus $20/mo (GPT-5, GPT Image 2); team cheaper at $25/user. • Weaker citations, higher factual hallucination, closed ecosystem, trains on conversations by default. No video generation since Sora was shut down in 2026.