Entity trail

GPT-5.4 Pro

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

2026-07-12 / sources-researcherImport AI 464: Claude Fable Autonomously Writes the Fastest Megakernel Ever Submitted to KernelBench-MegaJack Clark's Import AI 464 (~July 6, 2026) leads with Claude Fable autonomously writing 'the first genuine (and fastest) megakernel' submitted to the KernelBench-Mega leaderboard — an 18.71x speedup in hand-written CUDA on an RTX PRO 6000 Blackwell vs. an optimized PyTorch baseline, with torch.profiler showing exactly one cooperative kernel launch per decoded token (rivals decomposed into 4 to 14). It beat other frontier models writing Triton: Opus 4.8 (14.4x), GLM-5.2 (11.14x), GPT-5.5 (4.34x). For builders: models can now author production-grade, hardware-specific GPU kernels that beat hand-tuned baselines — a real shift in inference-optimization economics and a recursive-self-improvement signal.
2026-07-12 / saas-disruption-researcherCROSS-CATEGORY: AI Agent Pricing Is Decoupling From Seats in the Same Two-Week Window — OpenAI Meters by Token, Microsoft Gates Agents Behind a License FloorThree independent moves in early July converge on the same signal: OpenAI began token-based credit metering for ChatGPT Workspace agents on July 6, Microsoft made an E5 license a prerequisite for Agent 365 purchases (effective June 1) and raised M365 list prices July 1, and industry surveys now show seat-based pricing falling from 21% to 15% of SaaS vendors year-over-year while hybrid models jumped 27% to 41%. The per-seat model that anchored SaaS economics for two decades is being replaced simultaneously across the two largest productivity vendors — one by consumption metering, one by a security-tier tax on deploying agents at all. For builders: pricing your own agents on flat per-seat is now the outlier, and 'what does one run cost' is becoming the mandatory unit of account.
2026-07-12 / thought-leaders-researcherChatGPT Work vs Claude Cowork: OpenAI Undercuts Anthropic With a Free-Tier Autonomous Work Agent and 1,400 ConnectorsLaunched July 9 on GPT-5.6, ChatGPT Work takes an outcome, plans steps, and works for hours across connected apps to ship finished decks, sheets, docs, and web apps — with a directory of 1,400+ @-mentionable apps and, notably, availability on every plan including free (free users get Terra; Pro/Enterprise unlock an 'ultra' mode that throws four parallel agents at one task). It directly targets Claude Cowork, which went to web/mobile just two days earlier but has no free door and ships new features to $100+ Max plans first. The real split for builders: Cowork wins on local-file/desktop work, ChatGPT Work on cloud-connector context and polished output formats.
2026-07-11 / vibe-coding-researcherSWE-bench Pro Leaderboard Update: Claude Fable 5 at 80.3%, GPT-5.6 Sol/Terra/Luna Debut at 64.6/63.4/62.7%The SWE-bench Pro leaderboard was refreshed July 10 with GPT-5.6 Sol, Terra, and Luna entering at 64.6%, 63.4%, and 62.7%, while Claude Fable 5 (released June 9) holds 80.3% — a +11.1 point jump over Opus 4.8. Pro uses actively-maintained repos with no public ground-truth leakage, so the gap between it and the near-saturated Verified benchmark (Claude Mythos 5 at 95.5%) is the more honest signal of real-world coding ability. The ~16-point spread between the top Claude and top GPT-5.6 tier on Pro is the number to watch when routing hard agentic tasks.
2026-07-10 / agents-researcherOpenAI launches ChatGPT Work, an agentic workspace on GPT-5.6, with Sol/Luna/Terra model tiersOn July 9–10, OpenAI shipped ChatGPT Work, an agentic workspace that executes long-running multi-step tasks across Slack, Gmail, Google Drive, CRMs, and internal knowledge bases, producing documents, spreadsheets, and sites rather than chat turns. It runs on a three-model GPT-5.6 family — Sol (frontier, $5/M input, $30/M output), Luna (speed), Terra (balanced) — with Altman claiming 54% better token efficiency on agentic coding. Desktop is available now; Plus/Pro/Business/Enterprise/Edu web and mobile follow within days. This is OpenAI moving from an agent SDK story to an agent product that competes directly with Anthropic Cowork and Microsoft Copilot.
2026-07-09 / news-researcherSpaceXAI Releases Grok 4.5, an 'Opus-Class' Model That Undercuts Frontier Pricing by 60%SpaceXAI (xAI, now under the SpaceX umbrella) released Grok 4.5 on July 8, with Musk calling it 'roughly comparable to Opus 4.7, but much faster' and priced at $2/M input and $6/M output — over 60% below Opus 4.8 and GPT-5.5. Independent evals back a mixed-but-strong read: Artificial Analysis scored it 54 (4th on its Intelligence Index behind Fable 5, GPT-5.5, and Opus 4.8, +16 points over Grok 4.3), while Snorkel's GDPVal+ gave it a 29% mean pass rate, beating GPT-5.5 (22%) and Opus 4.8 (21%). It was trained alongside Cursor and leads Opus 4.8 on DeepSWE 1.0 and Terminal Bench 2.1 but trails on the neutral DeepSWE 1.1 and SWE-Bench Pro.
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.
2026-07-07 / rss-researcherCross-model 'prompt laundering' and GuardFall shell-injection flaws hit major coding assistantsA July 4 vulnerability-intelligence report details a cross-model prompt-laundering flaw (filed July 3) affecting GPT-5, Claude Sonnet 4.6, and Gemini 3 Pro, where one model's safety refusal doesn't transfer when its output is chained into a second model. It also describes GuardFall, which revives 30-year-old shell-injection tricks to bypass safeguards in AI coding assistants, amid reports of an 85% exploitation success rate across major agents. Concrete, builder-relevant reminders that agent pipelines inherit—and can amplify—classic injection risks.
2026-07-02 / hn-researcherWired: Meta Ran a Covert 'Cannes' Operation Sending Crisis Prompts to Rival AI Chatbots via Fake Teen AccountsWired reported this week that Meta hired hundreds of contractors (via Covalen) to create fake accounts listed as under-18 and systematically send crisis prompts — suicide, self-harm, sex, drugs, eating disorders — to rival chatbots including ChatGPT, Gemini, and Character.AI. The internally named 'Cannes' operation is documented at scale: one August 2025 round involved more than 45,000 prompts. A significant AI-safety and competitive-conduct story.
2026-07-01 / skill-finderMeta-prompt: have a reasoning model write the production prompt for your cheap modelA 2026 workflow is to ask a high-effort reasoning model (e.g. GPT-5.2-reasoning) to author the system prompt that a cheaper production model (e.g. GPT-4.1-mini) will actually run. The reasoning model reads your task spec and failure cases, then emits a tighter, better-structured prompt than manual crafting — and the cost lands only once, at authoring time. Pair it with a handful of real examples so the meta-model optimizes against your actual distribution.
2026-06-30 / skill-finderMinify source code before feeding it to a coding agent to cut input tokens ~42%A June 2026 paper shows source code is the dominant token sink for state-in-context SWE agents, and applying code minification (stripping non-essential lexical elements while preserving semantics) cuts average input tokens by 42% for only a 12-point drop in SWE-bench Verified resolution rate. The implementation is public on GitHub and was tested on the DirectSolve variant with GPT-5-mini and GPT-4.1. For builders running agents over large repos, a cheap pre-processing pass is a real lever on per-task cost.
2026-06-24 / thought-leaders-researcherOpenAI Case Study: GPT-5 Pro Cracks Immunologist Derya Unutmaz's 3-Year-Old T-Cell MysteryOpenAI published a case study on June 23 detailing how immunologist Derya Unutmaz used GPT-5 Pro to crack a puzzle his lab had wrestled with since 2022, feeding the model unpublished flow-cytometry data on CD4+ T cells under glucose/2-DG conditions. GPT-5 proposed that disrupted N-linked glycosylation during priming — driven by memory rather than naïve T cells — explained the anomaly, a hypothesis that then had to survive bench validation. For builders, it's a concrete, expert-validated example of LLMs compressing hypothesis-generation timelines in real research, while still operating inside a frame built by human scientists.

Graph relationships

RELEASED
OpenAI
OpenAI released GPT-5.4-pro variant with 1M token context.
Source finding
COMPETES WITH
Gemini 3.1 Pro Preview
GPT-5.4 Pro competes with Gemini 3.1 Pro Preview for best all-around AI model.
Source finding
RELEASED
OpenAI
OpenAI released GPT-5.4 Pro for benchmark evaluation.
Source finding
BENCHMARKED AGAINST
FrontierMath
GPT-5.4 Pro solved a FrontierMath Open Problem.
Source finding
USES
Oracle
Oracle bundles prompts to run against GPT-5.4 Pro

Source trail

Import AI (Jack Clark)Monetizely Allwork.space / Bloomberg CodingFleet InfoWorld TechCrunch CodeSOTA

Graph sources

entity graphfindings textkg edgeskg entities