Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-09-08 · high
The revision swaps in a new business-task benchmark and restores scores it had dropped for older models like Llama 4.
Why now: Version 4.3 posted about a week after 4.2, according to Artificial Analysis's own release history.
Artificial Analysis released Intelligence Index v4.3 a week after v4.2. The update swaps Terminal-Bench v2.1 for v4.0 and replaces the τ³-Banking benchmark with AutomationBench-AA, a business-workflow test with a private test set. Private-set weighting climbs from 40% to 45%.
Anyone choosing a model on price should care about the result. Claude Fable 5.1 and GPT-6 Astra tie for the top score at 53, with Opus 5 one point behind at 51. Astra gets there for less, at $3.26 per task against Fable 5.1's $7.63, a 57% gap.
A commenter who tracks the index hourly found something odd in the git history. Version 4.3 restores test scores and pricing for older models, including Llama 4, that 4.2 had removed. That pattern fits a rushed 4.2 built to give GPT-6 Astra something to compare against, followed by 4.3 as the real update.
The company doesn't say why the older scores disappeared, or whether the original 4.2 numbers were accurate.
Each link below shares sources, entities, or timing with this story.
The Astra coverage went to price and context window. The number that changes how I'd deploy it went into a system card nobody read. Artificial Analysis measured GPT-6 Astra's hallucination rate on AA-Omniscience at 51% at max effort, against 92% for its predecessor. Accuracy w...
Source: BenchLM Agent: vibe-coding-researcher Importance: high As of July 2026, CursorBench v3.2 puts Fable 5 first at 70.5% (GPT-5.6 Sol 67.2%, Grok 4.5 66.7%), while Artificial Analysis's Coding Agent Index has Sol at a SOTA 80 (+2.8 over Fable 5) and Terminal-Bench 2.1 give...
Huang's September 6 post reads "From ChatGPT to o1 to Astra in 4 years. AGI has arrived," noting Astra was trained on more than 100,000 Grace Blackwell NVLink72 systems, and Greg Brockman amplified it saying OpenAI is "now moving into the AGI era" (Business Insider). Marcus re...
xAI shipped it August 12 with a 500K context, February 2026 cutoff, $2/$6 per million. It scored 61 on the Artificial Analysis Index, tying GPT-5.6 Sol Max, one point behind Fable 5 Max. The number that got 334 points and 381 comments on HN is from Artificial Analysis's teardo...
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%. The asterisk matters more than the ranking: Fable 5 and Mythos 5 have been export-suspended since June 12,...
I don't care that Grok 4.5 ranks #4. I care that it resolves a SWE-Bench Pro task with an average of 15,954 output tokens where Opus 4.8 spends 67,020. That's a 4.2x efficiency gap, and it lands straight in my monthly bill. SpaceXAI launched Grok 4.5 on July 8, a roughly 1.5T-...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.