Du wirst angemeldet...

Bitte warte, während wir deine Anmeldung überprüfen

Artikel · Dienstag, 29. September 2026

AI developer tools · What shipped

For a senior engineer who already reads HN. Real changes in AI developer tools today: releases with version numbers, papers with benchmarks, repos that crossed a threshold worth knowing. Skip hype threads, pre-announcement leaks, and recycled summaries. Always link primary sources.

Von Marius BongartsTech81 Ausgaben
← Zur aktuellen Ausgabe
Ausgaben
9 / 81
Über Nacht von KI aus öffentlichen Quellen erstellt, täglich aktualisiert.
AI developer tools · What shipped
Dienstag, 29. September 2026
AI developer tools · What shipped

Claude Sonnet 5.5 ships; GPT-6 Luna reprices frontier tier

1 Min. Lesezeit

Claude Sonnet 5.5

Five variants, one benchmark leader.

Anthropic shipped Claude Sonnet 5.5 in September with intelligence scores reaching 56 on the Artificial Analysis Index—outpacing the frontier tier by sheer efficiency [Quelle: Artificial Analysis]. Output speeds hit 139 tokens per second across a 1M context window. Pricing spans $0.41 to $7.60 per task, letting teams dial in reasoning depth by workload.

The middle tier just became the production workhorse.

GPT-6 Luna repricing

OpenAI's six-model family resets frontier cost.

GPT-6 Luna ships with intelligence scores from 15 to 37, output speeds of 141–154 tokens per second, and pricing from $0.0045 to $0.07 per task [Quelle: Artificial Analysis]. The 1M token context window supports text and image inputs across Azure and OpenAI's API. A 15× pricing spread lets teams pick effort level by workload.

Frontier now means cost-per-capability, not cost-per-token.

Model release velocity

The lab race just hit one ship per day.

BenchLM tracked 215 notable AI model releases in the 12 months ending September 28, 2026—roughly one launch every two days [Quelle: BenchLM]. OpenAI led with 20 tracked releases; Alibaba followed with 17, Google with 14. August 2026 was the busiest month at 41 releases; 46% of classified launches were open-weight models.

Velocity favors vendors with established benchmark pipelines and rapid iteration loops.

Quellen
AI Model Release Statistics (2026): Launch Cadence by Lab
AI Model Release Statistics (2026): Launch Cadence by Lab
18 hours ago ... How fast are AI models shipping? Release counts, open-weight share, busiest months, and per-lab cadence from 212 tracked releases.
benchlm.ai
KI-Zusammenfassung

BenchLM tracked 215 AI model releases in the 12 months ending September 28, 2026, with roughly one notable release every two days. OpenAI led with 20 tracked releases, followed by Alibaba with 17 and Google with 14. August 2026 was the busiest launch month with 41 releases, and 46% of the 214 classified releases were open-weight models.

Quelle öffnen
Claude Sonnet 5.5: Release Intelligence, Performance & Price
Claude Sonnet 5.5: Release Intelligence, Performance & Price
10 hours ago ... The Claude Sonnet 5.5 release offers 5 models, each with different intelligence, performance, and pricing characteristics ... AI Agents · Evaluations.
artificialanalysis.ai
KI-Zusammenfassung

Anthropic released Claude Sonnet 5.5 in September 2026 with five model variants offering different intelligence-performance-cost tradeoffs. The top variant scores 56 on the Artificial Analysis Intelligence Index (incorporating 10 evaluations including SciCode, Humanity's Last Exam, and Terminal-Bench), outputs at 139 tokens/second, and costs $7.60 per task, while the most cost-efficient variant is $0.41 per task. All variants support a 1M token context window, reasoning capabilities, and text-image input.

Quelle öffnen
GPT-6 Luna Models - Intelligence, Performance & Price Comparison
GPT-6 Luna Models - Intelligence, Performance & Price Comparison
5 hours ago ... AI Trends · Premium · Log in. K. All Releases•. OpenAI logo. OpenAI. •. Proprietary release. •. Released September 2026. GPT-6 Luna: Release Intelligence, ...
artificialanalysis.ai
KI-Zusammenfassung

OpenAI released GPT-6 Luna in September 2026, a proprietary model family consisting of six variants with different intelligence, speed, and cost profiles. The top model (GPT-6 Luna max) scores 37 on the Artificial Analysis Intelligence Index v4.3.2, which incorporates 10 evaluations including AA-Briefcase, GDPval-AA, AutomationBench-AA, Terminal-Bench 4.0, SciCode, and others. The models support text and image inputs with a 1M token context window, with output speeds ranging from 141–154 tokens per second and pricing from $0.0045 to $0.07 per Intelligence Index task. GPT-6 Luna is available through Microsoft Azure and OpenAI's first-party API.

Quelle öffnen
Über Nacht zusammengestellt von MorningMail.aiZugestellt um 05:10