they are equally bad at estimating their own token cost, ask any agent to budget itself and it will happily plan 40 steps haha.
ECONOMY
-

Meta Launches Muse Spark 1.1 With Lower API Pricing
By
–
4/ Muse Spark 1.1 Meta shipped its first paid model API and priced Muse Spark 1.1 at $1.25 input and $4.25 output per million tokens, around a quarter of flagship rates. Zuckerberg posted on X for the first time in three years to launch it. It leads tool use benchmarks like MCP
-

GPT-5.6 Adds Tiered Pricing and Parallel Subagents
By
–
1/ GPT-5.6 (Sol, Terra, Luna) OpenAI split the generation into three tiers priced $5/$30, $2.50/$15, and $1/$6. Sol sets a new high on Agents' Last Exam at 53.6, and its ultra mode coordinates parallel subagents inside a single call. One catch worth knowing. METR logged its
-

Five New AI Models Compete on Cost and Capability
By
–
5 models dropped this week and price was the real battleground Two frontier labs shipped tiered families built to win on tokens per task, while the other three push past chat into playable worlds and clinical imaging. 1. GPT-5.6 Sol hits 88.8% on Terminal-Bench 2.1 and 92.2% on
-
Tech Giants Approach $700 Billion in AI Infrastructure Spending
By
–
-

GPT-5.6 Sol Leads Reported DeepSWE 1.1 Scores at Lower Cost
By
–
DeepSWE 1.1: GPT 5.6 Sol delivers the highest score at less than half the cost
-
An Earlier Analysis of AI and Robotics Risks to Chinese Employment
By
–
2/ My NYTimes op-ed on the potential job market and economic challenges of the robotic/AI revolution in China (NYTimes gift article): https://
nytimes.com/2015/06/11/opi
nion/chinas-troubling-robot-revolution.html?unlocked_article_code=1.xlA.0L6J.rZWIZAfhlI8U&smid=url-share
…
#RiseoftheRobots -
China Decode Discusses How AI and Robots Could Change Employment
By
–
https://
youtu.be/pdhDjcESt4o?si
=FhAIgY0ciEv2kpRt
… 1/ A very interesting discussion about the impact of #AI and #robotics on the job market in China between @Alicesqhan and @JKynge on the China Decode podcast. I actually wrote an op-ed for the @NYTimes on this, way back in 2015. I wrote: "China could -
OpenAI Claims GPT-5.6 Sol Uses Fewer Tokens Than Fable
By
–
GPT-5.6 sol is half the price and ~twice as token efficient as fable in many cases for accomplishing the same task. happy to deliver at one-quarter of the price.
-
Deterministic Tests Limit End-to-End Speed Gains From Faster AI Tokens
By
–
Fast mode isn't worth it. While token speed is faster, a good chunk of time goes on running of deterministic tests, hence your effective speedup is much smaller than advertised for 2x+ the cost. If you want to go faster, tell your agent to validate less, it would probably get