[KG] GPT-5.3 — risk

{
  "entity_slug": "gpt-5-3",
  "entity_name": "GPT-5.3",
  "entity_type": "ai_model",
  "title": "GPT-5.3: OpenAI's MoE Powerhouse Under Siege from Every Direction",
  "narrative": "GPT-5.3, OpenAI's Mixture of Experts flagship first observed February 2026, posts a GPQA of 92.0 and SWE-bench Pro of 56.8 at $1.75/$14 per million tokens. It deploys RLHF, Chain-of-Thought, and Sparse MoE — but faces a four-front competitive war. Claude Mythos Preview, Claude Opus 4.7, GPT-Rosalind, and even the legacy GPT-3.5 all claim rival status. The model runs on Qualcomm hardware and powers Computer Use, yet its own successor GPT-5.5 already tops benchmarks at double the API cost. Agent Cloud, Anthropic, King's College London, and Codex CLI all use GPT-5.3, signaling broad adoption. But with GPT-5.4 failing banking benchmarks and GPT-5.5 launching as a super-app strategy, GPT-5.3 sits in a precarious middle — strong on paper, but already leapfrogged internally and externally.",
  "key_points": [
    "Competes with four models: Claude Mythos Preview, Claude Opus 4.7, GPT-Rosalind, GPT-3.5",
    "Deploys RLHF, Sparse MoE, and Chain-of-Thought prompting",
    "Runs on Qualcomm hardware; powers Computer Use product",
    "Successor GPT-5.5 already launched with higher costs and mixed results",
    "Adopted by Anthropic, Agent Cloud, and Codex CLI despite internal competition"
  ],
  "angle": "risk",
  "neighborhood_size": 19,
  "generated_at": "2026-06-21T22:50:10.223731+00:00"
}