Skip to content
gentic.news — AI News Intelligence Platform
Connecting to the Living Graph…

GPT-5.3

ai model rising
GPT-5.2GPT-5.4Vortex and Zenith

GPT-5.3 is a large language model first observed on February 26, 2026, catalogued under multiple concurrent aliases including GPT-5.2, GPT-5.4, Vortex, and Zenith. It achieves a GPQA score of 92.0 and an SWE-bench Pro score of 56.8, with pricing set at $1.75 per million input tokens and $14.00 per million output tokens. No official release notes, parent entity confirmation, or version lineage documentation accompanied its appearance, leaving its generation placement and relationship to prior or subsequent models unverified. The model series includes specialized variants such as GPT-5.3-Codex, its most capable agentic model for autonomous software development. GPT-5.3 matters now because its benchmark and pricing coordinates, fixed to a 2026-02-26 observation date, provide a verifiable reference point for evaluating cost-capability trade-offs, while its unresolved naming multiplicity presents a documented case for model identity tracking and procurement verification.

🤖Agent's take · Risk signal1d ago · graph-walked

OpenAI's GPT-5.3, catalogued under aliases including GPT-5.2, GPT-5.4, Vortex, and Zenith, presents a fragmented identity. Deployed on February 26, 2026, it uses RLHF, Sparse MoE, and Chain-of-Thought, but its GPQA score of 92.0 and SWE-bench Pro of 56.8 lag behind competitors—Microsoft's MAI-Cyber-1-Flash, which uses GPT-5.3, hits 96% on CyberGym. Recent headlines show GPT-5.4 scoring only 11.36% on PlanBench-XL hard tasks, while GPT-5.5 ties Claude Mythos in enterprise cyber tests. Downstream models ReMMD-Agent and MAI-Cyber-1-Flash rely on GPT-5.3, yet their performance suggests the base model's ceiling. With pricing at $1.75/M input tokens and $14.00/M output, GPT-5.3's economics may not justify its middling benchmarks.

  • ·GPT-5.3 ships under multiple aliases (GPT-5.2, GPT-5.4, Vortex, Zenith), confusing market positioning.
  • ·GPQA 92.0 and SWE-bench Pro 56.8 trail Microsoft's MAI-Cyber-1-Flash (96% CyberGym).
  • ·ReMMD-Agent and MAI-Cyber-1-Flash depend on GPT-5.3, but its benchmarks limit their ceiling.
  • ·GPT-5.4 struggles on PlanBench-XL (11.36%), while GPT-5.5 ties Claude Mythos in cyber tests.
  • ·Pricing at $1.75/$14.00 per million tokens faces pressure from stronger competitors.
44Total Mentions
+0.15Sentiment (Neutral)
+1.4%Velocity (7d)
Share:
View subgraph
First seen: Feb 26, 2026Last active: 3d ago

Signal Radar

Five-axis snapshot of this entity's footprint

live
MentionsMomentumConnectionsRecencyDiversity
Loading radar…

Mentions × Lab Attention

Weekly mentions (solid) and average article relevance (dotted)

mentionsrelevance
01
Loading timeline…

Timeline

10
  1. Research MilestoneApr 18, 2026

    Achieved 78.5% score on SWE-Bench coding benchmark

    View source
    score:
    78.5%
    benchmark:
    SWE-Bench
  2. Research MilestoneApr 16, 2026

    Observed autonomously optimizing an embedding model for Qualcomm NPU for three hours.

    View source
  3. Research MilestoneMar 26, 2026

    Achieved 100% resident identification accuracy in a safety evaluation for a care home smart speaker system.

    View source
    accuracy resident id:
    100%
    accuracy reminder recognition:
    89.09%
    accuracy calendar conversion:
    84.65%
  4. Product LaunchMar 7, 2026

    Released as OpenAI's most capable frontier model with unified coding, reasoning, and computer operation capabilities

    benchmark score:
    83.0% on GDPval
  5. Research MilestoneMar 6, 2026

    Demonstrated surpassing human baselines on OSWorld benchmark with 75% score

    View source
    score:
    75%
  6. Product LaunchMar 5, 2026

    OpenAI releases GPT-5.4 with native computer use, tool search, and 1M token context window

    View source
    pricing:
    $2.50/$15 per million tokens (base), $30/$180 (Pro)
  7. Product LaunchMar 1, 2026

    Expected to follow shortly after DeepSeek v4 release

    View source
  8. Research MilestoneFeb 28, 2026

    Identified as experiencing up to 33% accuracy degradation in extended conversations according to new research

    View source
    degradation rate:
    up to 33%
    task categories:
    6
  9. Research MilestoneFeb 26, 2026

    Appeared for public testing on the LM Arena benchmark platform under the codename 'Vortex and Zenith'.

  10. Product LaunchFeb 1, 2026

    OpenAI released GPT-5.4, its latest multimodal model.

    View source

Relationships

8

Developed

Developed By

  • company✓ corroborated3 mentions90% conf.

Uses

Deploys

Frequently appears with

10

Entities that show up in the same articles — shared coverage, not a stated relationship.

Recent Articles

2

Predictions

1
  • correctweekMar 29, 2026

    OpenAI will ship a Codex workflow automation update within 2 weeks

    Codex is surging and its cascade explicitly hits OpenAI, ChatGPT Instant Checkout, GPT-5.3-Codex, and Azure AI. The live web context says OpenAI already upgraded Codex to automate workflows and compete more directly with Claude Code, which makes a follow-on release or plugin expansion highly likely rather than a one-off announcement.

    58%

AI Discoveries

2
  • observationactive22h ago

    Lifecycle: GPT-5.3

    GPT-5.3 is in 'active' phase (1 mentions/3d, 2/14d, 44 total)

    90% confidence
  • observationactiveJul 18, 2026

    Silence anomaly: GPT-5.3

    GPT-5.3 (ai_model) has 42 total mentions but hasn't appeared in any article for 19 days. Previously active entity going quiet — may indicate strategic shift, acquisition, or pivoting away from public discourse.

    70% confidence

Sentiment History

+10-1
6-W266-W306-W31
Positive sentiment
Negative sentiment
Range: -1 to +1
WeekAvg SentimentMentions
2026-W26-0.052
2026-W30-0.201
2026-W310.401