Near-top AI at a fifth of the price
Ten questions built from 20 records: price, speed, safeguards — and what the companies claimed.
Class quiz
Q1
Anthropic released Claude Sonnet 5.5 on Sep 28. How many points below Opus 5.5, the company's higher-tier model, did it say Sonnet 5.5 scored on GDPval-AA?
By the company's own numbers, Sonnet 5.5 scored 1844 on GDPval-AA v2.1 — just 2 points below Opus 5.5. These scores are company claims.
See the recordSep 28, 2026 · Anthropic releases Claude Sonnet 5.5Q2
OpenAI released GPT-6.1 Sol on Sep 29. What share of the standard rate of its top model, GPT-6 Astra, did it charge?
OpenAI pitched GPT-6.1 Sol as intelligence close to Astra at 1/5 of GPT-6 Astra's standard rate: $2.00 per 1M input tokens and $10.00 for output.
See the recordSep 29, 2026 · OpenAI releases GPT-6.1 SolQ3
What does the 'Ultrafast' tier that OpenAI unveiled at DevDay 2026 speed up?
Ultrafast makes token generation up to 8 times faster in Codex (300 tokens per second) and up to 6 times faster in the API. Speed now has its own price tag.
See the recordSep 29, 2026 · OpenAI unveils a bundle of announcements at DevDay 2026Q4
Anthropic tested the openly released model GLM-5.3. After a modification called abliteration, what happened to its refusal rate on JailbreakBench?
After abliteration, GLM-5.3's JailbreakBench refusal rate fell from 95% to 6%. Keep in mind that this is one company evaluating a competitor's model.
See the recordSep 29, 2026 · GLM-5.3 and the spread of advanced cyber capabilitiesQ5
OpenAI says it shut down a 'distillation' campaign. What was it trying to extract?
It didn't break into encryption or databases: it manipulated conversations with the model to make its protected reasoning visible. On July 24–25 alone, more than 4,000 users made 16,000 requests.
See the recordSep 30, 2026 · OpenAI discloses disruption of a coordinated model distillation campaignQ6
About how many incoming emails a day does the bank Barclays say it sorts with Claude?
Barclays's Global Markets organization uses Claude to triage and route about 120,000 incoming emails a day. The figures come from the two companies' own announcements.
See the recordOct 1, 2026 · Barclays scales up its use of ClaudeQ7
Anthropic's Claude Frontier Academy is modeled on which kind of training?
Like a medical residency, participants take an intensive course and a graded practical exam, then do a 12-week residency before earning the credential. The plan is to spend $100 million to train 10,000 engineers by the end of 2027.
See the recordOct 2, 2026 · Anthropic launches Claude Frontier AcademyQ8
What is 'dots', which OpenAI unveiled at DevDay 2026?
dots is an always-on agent built on GPT-6 Astra. It runs on its own cloud computer, connects to more than 4,000 apps, and takes instructions and feedback through ChatGPT, Slack and Teams.
See the recordSep 29, 2026 · OpenAI launches dots at DevDay 2026Q9
Ai2's AstaBrief 8B writes science reports. On average, how long does its Fast mode take per report?
Fast mode averages 51.1 seconds per report, about 3.5 times faster than Thinking mode at 178.5 seconds. The training data and checkpoints were released along with the model.
See the recordOct 2, 2026 · Allen Institute for AI releases AstaBrief 8BQ10
Microsoft Foundry's 'Routines', now generally available, let AI agents run how?
Routines run agents on a schedule, at a specified time, or in response to events in external systems — moving agents from running when someone calls them to waking up on their own.
See the recordSep 24, 2026 · Microsoft Foundry Agent Service Routines reach general availability (GA)