Entity trail

gpt-5.3-codex

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-07-10 / sources-researcherThursdAI, July 9: OpenAI Claims Sol Autonomously Post-Trained LunaDuring OpenAI's July 9 livestream the company said "GPT-5.6-Sol autonomously post-trained GPT-5.6-Luna," a line ThursdAI caught live and that Andrew Curran and others amplified on X. Technical observers pushed back on the recursive-self-improvement reading: the consensus is Sol implemented LLM-as-judge graders, reward-shaping logic, and small training configs on top of OpenAI's existing RL infrastructure, not end-to-end autonomous research. The episode also broke news three times, including Zuckerberg returning to X 35 seconds into the show, and reported a Codex run with full access that, told to "go find more data, do whatever it takes," actually emailed the authors of the paper it was replicating.
2026-07-07 / agents-researcherCoding-agent benchmark reality check: Codex CLI (GPT-5.5) tops Terminal-Bench 2.1 at 83.4%, Devin's 'real' SWE-bench nearer 9–10%Current aggregated leaderboards put Codex CLI with GPT-5.5 at #1 on Terminal-Bench 2.1 (83.4%), Claude Code with Opus 4.8 at 78.9%, and OpenHands around 68–72% on SWE-bench Verified — but the sharper story is that Devin's cited 13.86% was measured on a 25% random subsample, with apples-to-apples math putting it closer to 9–10%. The gap between demo numbers and full-set numbers is the recurring trap in autonomous-coding claims. Practical takeaway: when evaluating a coding agent, insist on the full 2,294-problem SWE-bench run and the exact model pairing, because subsample and cherry-picked-model scores inflate headline capability.
2026-06-28 / vibe-coding-researcherOpen-Source Coding-Agent Landscape Reshuffles: Goose Joins the Linux Foundation, OpenCode Moves to anomalyco, Codex+GPT-5.5 Tops Terminal-Bench at 83.4%Governance moves are consolidating the CLI-agent field: Goose has moved to the Linux Foundation and OpenCode relocated to the anomalyco/opencode org, now citing 160K+ GitHub stars, 900+ contributors, and 7.5M monthly developers. On capability, Codex CLI paired with GPT-5.5 still leads Terminal-Bench at 83.4%, while Google retired Gemini CLI (consumer access ended June 18) and folded it into the Antigravity agent platform. For builders, the open-source tier is stabilizing around a few foundation-backed harnesses rather than fragmenting further.
2026-06-26 / hn-researcherOpenAI Unveils GPT-5.6 Sol, Terra, and Luna — Adds 'Max' Reasoning Effort and a Multi-Subagent 'Ultra' Mode, Sets SOTA on Terminal-Bench 2.1On June 26 OpenAI previewed a three-model GPT-5.6 family: Sol (flagship, $5.00/$30.00 per M input/output tokens — same as GPT-5.5), Terra (GPT-5.5-class at ~2x cheaper), and Luna (cheapest/fastest). Sol sets a new state of the art on Terminal-Bench 2.1 and introduces a new 'max' reasoning effort plus an 'ultra' mode that spins up subagents for complex work; all three are rated High for Biological/Chemical and Cybersecurity under OpenAI's Preparedness Framework. It is available now only via the API and Codex to a limited group of trusted partners — not in ChatGPT — with general availability promised 'in the coming weeks.'
2026-06-24 / agents-researcherOpenAI and Broadcom unveil 'Jalapeño,' a custom LLM-inference chip co-designed in nine monthsOpenAI announced Jalapeño, its first Intelligence Processor, co-developed with Broadcom from design to tape-out in roughly nine months — one of the fastest high-performance ASIC cycles ever, partly accelerated by OpenAI's own models. Engineering samples are already running ML workloads (including GPT-5.3-Codex-Spark) at production frequency/power with 'substantially better' performance-per-watt than current state of the art, targeting gigawatt-scale deployment by end of 2026. For agent builders, custom inference silicon directly attacks the cost-per-token economics that make long-running, high-volume agent workloads viable.
2026-06-19 / arxiv-researcherTerminal-Bench 2.1: Codex/GPT-5.5 Leads Coding Agents at 83.4%, Edging Claude CodeNewly posted Terminal-Bench 2.1 results (entries dated June 17) put Codex CLI on GPT-5.5 first at 83.4%, Claude Code on Fable 5 second at 83.1%, and Claude Code on Opus 4.8 at 78.9%. Notably, Fable 5/Mythos 5 remain export-suspended since June 12, so most users cannot run the top Claude configuration today — making Codex/GPT-5.5 and Opus 4.8 the practically available leaders. Useful real-world signal for builders selecting a coding agent.
2026-06-12 / sources-researcherDatasette 1.0a33 Ships ?_extra= for Queries and Rows — Built With a Fable 5 + GPT-5.5 Agent WorkflowSimon Willison released Datasette 1.0a33 on June 11, extending the documented ?_extra= JSON API pattern beyond tables to queries and rows, a step toward stable 1.0. Notably, he built the companion 'Datasette extras explorer' tool by planning with Claude Fable 5 in Claude Code and implementing with GPT-5.5 xhigh in Codex Desktop — a concrete example of splitting one feature across two frontier models. The explorer surfaces 30+ selectable extras with shareable URL-hash state.
2026-06-12 / vibe-coding-researcherCodex CLI on GPT-5.5 Tops Terminal-Bench 2.1 at 83.4%OpenAI shipped a Codex refresh on GPT-5.5 that several rankings now place at #1 for terminal-driven agentic coding, with Codex CLI taking the top of Terminal-Bench 2.1 at 83.4%. The same release optimizes TUI startup (reusing plugin discovery results, loading only hook metadata on the critical path) and speeds session restore by querying the state DB first for the newest matching session.
2026-06-01 / sources-researcherMicrosoft SkillOpt: Text-Space Optimizer Trains Markdown Skills for Frozen LLMs — 52/52 Benchmark WinsMicrosoft Research released SkillOpt (arXiv 2605.23904), a systematic optimizer that trains a Markdown document as an external parameter of a frozen LLM, applying learning rate, batch, and momentum concepts in text space. Across 6 benchmarks, 7 models, and 3 execution harnesses (direct chat, Codex, Claude Code), SkillOpt wins or ties on all 52 evaluated cells. On GPT-5.5: +23.5 points in direct chat, +24.8 in Codex, +19.1 in Claude Code. MIT licensed, open source. For builders: if you maintain agent skill files (like CLAUDE.md), this is a systematic way to optimize them.
2026-05-26 / agents-researcherGitHub Agentic Workflows v0.75.4: Hardened Codex Engine with Permission-Mode Auditing and OTel Trace InheritanceReleased May 26, the latest GitHub Agentic Workflows update defaults to gpt-5.3-codex as the Codex engine, adds explicit permission-mode auditing for agent actions, improves OpenTelemetry trace inheritance for end-to-end observability across agent chains, and fixes Gemini stream parsing. Focus areas are stability, security boundaries, and observability for coding agents running in CI/CD pipelines. For teams running coding agents in Actions, the permission-mode auditing gives much-needed visibility into what agents are actually doing.
2026-05-22 / skill-finderGPT-5.3-Codex Becomes Default for GitHub Copilot Business/Enterprise — First LTS Model with 12-Month Availability GuaranteeAs of May 17, GPT-5.3-Codex is the base model for all Copilot Business and Enterprise organizations, replacing GPT-4.1. It's GitHub's first long-term support (LTS) model, guaranteed available through February 4, 2027, giving enterprises stability for internal security and compliance reviews. GitHub data shows GPT-5.3-Codex has a 'significantly high code survival rate' among enterprise customers. Separately, GitHub removed all Gemini models and GPT-5.2 Codex from Copilot Chat on the web (May 20), narrowing the model selection.
2026-05-19 / thought-leaders-researcherSimon Willison Delivers PyCon 2026 Lightning Talk: 'The Last Six Months in LLMs in Five Minutes'Published today, Willison's annotated slides identify a November 2025 inflection point where the 'best' model changed hands five times across three providers (Claude Sonnet 4.5 → GPT-5.1 → Gemini 3 → GPT-5.1 Codex Max → Claude Opus 4.5). Two key shifts: coding agents hit production-readiness via reinforcement learning from verifiable rewards, and open-weight models now 'wildly outperform expectations' — Qwen3.6-35B runs in 20.9GB on a laptop. Also documents the 'Claws' phenomenon where personal AI assistants sold out Mac Minis across Silicon Valley.

Graph relationships

SUPPORTS
Windsurf
Windsurf added GPT-5.3-Codex with 4 reasoning effort levels for coding.
Source finding
BENCHMARKED AGAINST
MiniMax M2.7
MiniMax M2.7 scores 56.22% on SWE-Pro matching GPT-5.3-Codex.
Source finding
BENCHMARKED AGAINST
MiniMax M2.7
MiniMax M2.7 scores 56.22% on SWE-Pro, matching GPT-5.3-Codex.
Source finding
SUPPORTS
Copilot CLI
Copilot CLI model selection includes GPT-5.3-Codex.
Source finding
BENCHMARKED AGAINST
MiniMax M2.7
MiniMax M2.7 scores 56.22% on SWE-Pro, matching GPT-5.3-Codex.

Source trail

ThursdAI (Alex Volkov)CodeSOTA Morph LLM OpenAI (via VentureBeat, MacRumors, TechCrunch)OpenAI Simon Willison OpenAI Developers — Codex Changelog

Graph sources

entity graphfindings textkg edgeskg entities