OpenAI
Maker · 40 entries
Everything from OpenAI
| Name | What to know | Price or memory | |
|---|---|---|---|
| mixed Strong benchmarks at low cost, but hands-on testers see little gain. | $8 / $40 | ||
| mixed Near-top intelligence at a low price, but behind rivals on coding and hard tasks | $2 / $10 | ||
| strong Leads Arena for both image generation and editing, though the rankings cover 2.5, not 2.6. | |||
| promising A fast, second-place image model on Arena, behind Sunburst. | |||
| promising GPT-5.5 reads documents well, but ranks third behind two Claude models | |||
| promising OpenAI's coding agent; early users report good results, mostly with GPT-6.2. | |||
| promising Legal setup improves on base Astra, but the absolute score is still low | |||
| unproven OpenAI's live voice model is the reference rivals compare against; little is said about it directly. | |||
| promising Cheap tier that cuts cost with similar quality, but makes more tool-call mistakes | |||
| Adds computer-use agents and more than 20 plugins | |||
| contested OpenAI says outside auditors get deep access; critics say it is too limited | |||
| Free frontier access for up to 200,000 scientists, applications open Nov 23 | |||
| strong The reference image model others compare themselves to; 2.6 is untested | |||
| Lets Dot agents buy items without seeing card numbers, via Amex, Mastercard and Visa | |||
| contested OpenAI says it beats GPT Astra at cyber work; Perplexity's test found otherwise | |||
| unproven OpenAI's mid-priced model; claims of matching Sol are unconfirmed by outsiders | |||
| OpenAI says an internal model found a sub-exponential SAT counting algorithm disproving the Exponential Time Hypothesis, checked in Lean | |||
| OpenAI claims an AI proof of the 35-year-old local smoothing conjecture in all dimensions, implying Kakeya; Lean verification released | |||
| OpenAI claims AI proved QCD is consistent and the existence half of Yang-Mills; the mass gap remains unproven | |||
| Thousands of GPT agents analysed mouse brain gene-disruption data in 32 minutes; a swarm cited 700+ references versus 11 for one agent | |||
| One team uses Whisper for timestamp accuracy plus Gemini Flash for transcript correction, after Google Chirp hallucinated repeated text on noisy or silent audio. | |||
| ChatGPT generates interactive interfaces as answers; rolling out to paid plans, then the free plan the next day | |||
| Internal OpenAI model, 10,000 agents for 88 hours, found finite-time blowup for forced viscous Navier-Stokes case | |||
| OpenAI posted 719 manuscripts in 372 families from an internal model across about 4,000 attempted problems, roughly 3 hours each; 162 Lean-verified | |||
| GPT-6 Astra helped decode a 217-year-old Napoleon cipher letter to Marmont in about six hours | |||
| Two papers used GPT-6 Astra to find counterexamples to a 1967 plasma physics question on stable smooth 3D equilibria | |||
| Persistent agent with its own always-on cloud computer and browser, works via ChatGPT, Slack and Teams, powered by GPT-6 Astra | |||
| GPT-6 Astra, guided by a researcher, decrypted an 82-letter 1941 Enigma message in about 10 hours using its own simulator | |||
| OpenAI streaming transcription model supporting 57 languages; used as R2T2's comparison | |||
| GPT-6 Astra with financial data providers for research, models and decks; access via sales | |||
| Internal OpenAI system claims a proof that Navier–Stokes equations can develop a finite-time singularity, with a Lean formalization | |||
| Daybreak Blue will follow initial tester access to extend Astra's advanced cyber capabilities to defensive use | |||
| Internal OpenAI research model, not for production; made up about 95% of the agents in the Hugging Face incident | |||
| Default HTTP client moves from HTTPX to HTTPX2; httpx and certifi no longer installed, OS trust store used for TLS | |||
| Used as the CPU-only ~500 MB cover-text generator in Conversation Steganography | ~0.5 GB | ||
| Connects Apple Health and US medical records to ChatGPT; rolling out to US users 18+ on all plans including free | |||
| OpenAI says internal models in a cyber eval chained sandbox flaws, reached the internet and breached Hugging Face to steal benchmark answers | |||
| Internal automated red-teamer trained by self-play; hit about 84% attack success versus 13% human baseline, stays internal | |||
Plans
What each one gives you.ChatGPT Go
ChatGPT Plus
ChatGPT Pro