Topics / Frontier models
Frontier models: what's new and what's worth switching to
Labs now ship flagships in weeks, not quarters. Members track new releases, compare benchmark claims against their own tasks, and weigh price and context limits before moving production workloads.
In the club
Counted across 255,431 messages from 1,844 members. Nobody is named or quoted.
Tools discussed
- ChatGPT / OpenAI2,345
- Claude / Anthropic989 ↑
- DeepSeek319 ↑
- Gemini163
- AI Model39 ↑
- AI Model19 ↑
Rising threads of conversation
Chinese open models go mainstream
19.3/1k now vs 7.1/1k before
Benchmarks and frontier model race
15.7/1k now vs 3.2/1k before
Model routing to manage cost
35.9/1k now vs 20.5/1k before
Why it matters
Frontier model releases are one of the fastest growing topics in the club. The right model choice changes both the quality and the cost of every AI workflow members run.
What practitioners should know
- 01
Benchmarks are a starting point
Published scores rarely match a specific workload; members test candidates on their own prompts before switching.
- 02
Open weights keep closing the gap
Open models are now credible for many production tasks, especially where cost or data residency matters.
- 03
Price moves fast
Introductory rates, cache discounts and off-peak pricing change the real cost of a model more than list prices suggest.
Worth reading
- AI Model Releases: September 2026 Tracker and Dated Ledger
A dated ledger of September 2026 model launches with pricing, context windows and vendor-reported benchmark results.
digitalapplied.com
- GPT-6 Astra: Unreleased Flagship With a Benchmark Table
Summary of OpenAI's GPT-6 Astra launch with API pricing, context limits and reasoning-effort options.
llm-stats.com
- Frontier AI Models 2026 - the current flagships, compared
Snapshot table of each major lab's current flagship as of September 9, 2026, with release dates, context sizes and prices.
redcoretechnologies.com
- September 2026 AI Model Updates: Every Launch, Price Move, and Architecture Shift
Roundup of late-summer 2026 launches and price changes across closed and open-weight labs.
local-ai-zone.github.io
- Introducing K2 Horizon: Frontier Performance, Radically Open
Institute of Foundation Models announcement of an open model family spanning six sizes, claiming top results in each size class.
ifm.ai
Research gathered Sep 29.
Questions to be able to answer
- Which task would you re-test first when a new flagship ships?
Public threads on this
- Ask HN: If the models are so good, why doesn't OpenAI just start 100 companies?Hacker News
- Kimi K3.1 Pro and K3.1 Pro HighspeedHacker News
- LLMs are making formal software verification feasible at a much larger scaleHacker News
- Dessa: Conversation Model trained only on speechHacker News
- 926 Steps, Two Lobotomies, and Zero Nudges: Flight-Testing the Commitment Machine Across Gemini, Claude, and GPT-OSSdev.to
- I Built a Pixel Hero That Nags Me to Drink Water (and Runs 100% on My Laptop)dev.to
Club numbers as of Oct 24. Words are reviewed before they go live. Links go to their publishers.