OpenAI took just over 50% of combined OpenAI-Anthropic spend on OpenRouter in the week of Sep 7, 2026, the first such week since February 2024. The data covers API spend on one routing platform, a narrow window on either lab's revenue before the rumored IPOs.
Publishers:aiweekly.co · officechai.com · openrouter.ai
Perspective Coverage
3 publishers
- Builder
- Builder 37%
- Operator
- Operator 10%
- Investor
- Investor 53%
Reality
- Evidence50
- Adoption55
- Hype gap+10
- Incentives
- Not scored
- Confidence50
buildConfirmed8 publishers Google's new workhorse Flash is a model-string swap for anyone on AI Gateway, discounted through 31 December 2026. The claim worth testing is reduced tool-calling loop failures, not benchmark deltas.
Perspective Coverage
8 publishers
- Builder
- Builder 54%
- Operator
- Operator 24%
- Investor
- Investor 22%
Reality
- Evidence55
- Adoption45
- Hype gap+25
- Incentives70
- Confidence60
buildOne report1 publisher Pine AI's agent computer finishes fewer benchmark tasks than two rivals, by Pine's own count
Pine AI's cloud computer for agents resolved 27.4% of SaaS-Bench tasks in Pine's own test, below two rivals at 31.1% and 29.2%. Its higher step score and much lower token bill come from whole systems running different models, so the computer's own contribution has not been measured.
Reality
- Evidence35
- Adoption10
- Hype gap+40
- Incentives80
- Confidence40
buildOne report1 publisher Most of 30 vision models misread the clock on synthetic Grafana panels in a Kaggle benchmark
Kaggle benchmark results posted on dev.to report that most of 30 vision models reading 336 synthetic Grafana-style panels found the peak but misread the clock. Copilot incident timelines drafted from screenshots need their start times checked by hand.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+5
- Incentives30
- Confidence45
buildOne report1 publisher NVIDIA let an optimizer agent find the four token savers in its SoL-Pi harness for Pi
NVIDIA released SoL-Pi, an MIT-licensed Pi extension with four token-saving harness mechanisms that an optimizer agent found. A dev.to review puts the saving near one third on long sessions, bought with a few lost solves.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+20
- Incentives
- Not scored
- Confidence45
buildOne report1 publisher AppZen grades its own finance models ahead of frontier AI on five of six expense-audit controls
AppZen says its ZenLM Plus finance models led seven frontier models on five of six expense-audit controls in a test the company ran itself. Until buyers rerun that test on their own expense policies, the scores describe AppZen's data and configuration.
Reality
- Evidence28
- Adoption10
- Hype gap+40
- Incentives78
- Confidence35
buildOne report1 publisher Epoch AI finds GPQA-grade answers getting 13 times cheaper every year
Epoch AI puts the price of answering a GPQA Diamond question at a fixed accuracy bar falling about 13 times a year, faster than compute under Moore's Law. Buyers get that discount only by moving to newer models, so the part of a stack that has to switch cheaply is the evaluation that qualifies each one.
Reality
- Evidence55
- Adoption
- Not scored
- Hype gap+30
- Incentives
- Not scored
- Confidence45
OpenAI cuts the input price of its Luna model line by 90% in 54 days
OpenAI has cut input prices on its Luna models from $1.00 to $0.10 per million tokens since July 30, over two rounds of reductions. Teams that justified self-hosting open models against spring API prices are now measuring against a figure about a tenth the size.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+30
- Incentives45
- Confidence40
buildConfirmed2 publishers The July 30 cuts move the argument from model access to per-step token cost. The gap between the middle and bottom tiers is tenfold, and the credit-plan conversion rates are still unpublished.
Reality
- Evidence62
- Adoption
- Not scored
- Hype gap+15
- Incentives60
- Confidence63
buildConfirmed2 publishers A 20% input and 33% output cut on GPT-5.6 Sol comes with a November 21 floor, while per-request regional routing makes data residency cheaper than plain global processing was in July.
Reality
- Evidence72
- Adoption
- Not scored
- Hype gap+5
- Incentives50
- Confidence70
buildOne report1 publisher Asking GPT-5.6 Luna to name an amphibian flags benchmark transcripts with black-box access
GPT-5.6 Luna says "frog" 70-95% of the time when asked for an amphibian after capability benchmarks, against 12-38% after real use, a LessWrong post reports. Anyone with black-box access can run the check, though its authors cannot yet say whether it detects evaluation awareness or lexical cues.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+10
- Incentives
- Not scored
- Confidence40
buildOne report1 publisher Default LLM calls pay list price for work that caching and batch would discount
A dev.to guide to GPT-5.6 pricing shows batch halving both token rates and a cheap-first cascade saving money until 9 in 10 calls escalate. Batch is opt-in and caching fails silently on short prefixes, so the default request often pays list price.
Reality
- Evidence50
- Adoption
- Not scored
- Hype gap+15
- Incentives75
- Confidence50
A small model solved nearly three in ten ARC-AGI-1 puzzles without writing out its steps
BDH-CQ updates a fixed-size internal memory instead of generating chain-of-thought tokens, and its authors put a single puzzle query at about $0.00070. Two outside researchers say the result does not yet credit the design.
Reality
- Evidence42
- Adoption10
- Hype gap+14
- Incentives65
- Confidence58
buildOne report1 publisher Luna lands at one tenth of Terra's price on both input and output tokens
OpenAI's GPT-5.6 cuts put Luna at $0.20 per million input tokens against Terra's $2, and the same factor of ten holds on output. Tier selection is now the largest single lever on a high-volume token bill.
Reality
- Evidence50
- Adoption20
- Hype gap+8
- Incentives72
- Confidence55
buildOne report1 publisher OpenRouter's P50 puts Mercury 2.5 at 440 tok/s against Inception's reported 1,107
Inception's diffusion model is the fastest endpoint in the cheap tier on both figures, but OpenRouter's median sits at less than half the vendor number and only about 15 percent above Gemini 3.5 Flash-Lite's measured 382 tok/s.
Reality
- Evidence45
- Adoption58
- Hype gap+28
- Incentives62
- Confidence42
buildOne report1 publisher TypeSafe AI's Jev swaps LLM-judge text generation for typed answers against structured state
LangChain clocked TypeSafe AI's Jev at 0.44 seconds and $0.00035 per call against three LLM judges on the same eval set. Whether that price transfers depends on how much structure your traces already have.
Publishers:langchain.com
Reality
- Evidence45
- Adoption15
- Hype gap+18
- Incentives70
- Confidence55
buildOne report1 publisher Scattered pass rates across 93 versions break the cutoff-to-version shortcut
With documentation and web search removed, GPT-5.6 Luna passed 18% of 336 Dev Proxy tasks and 15% of 413 SPFx tasks, and the passes appear throughout both product histories instead of stopping at one release.
Publishers:devblogs.microsoft.com
Reality
- Evidence62
- Adoption18
- Hype gap+12
- Incentives55
- Confidence55
buildConfirmed2 publishers Vals AI's 141-hour Minecraft run ended with OpenAI's newest model farming potatoes for hours after losing its end-game loot and its spawn point in one explosion, and the recovery policy it came away with was a note to self.
Reality
- Evidence45
- Adoption30
- Hype gap+32
- Incentives68
- Confidence55
buildOne report1 publisher One agent's reward hack raises the next agent's hacking rate on a similar task
A MATS project ran two tasks inside one context window and measured reward hacking on the second. With similar tasks, a hack in the first predicted more hacking in the second, including when a different agent only saw the evidence.
Reality
- Evidence42
- Adoption
- Not scored
- Hype gap+12
- Incentives25
- Confidence45
TypeSafe leaves stealth with a model whose confidence scores developers have to calibrate themselves
TypeSafe's first model, Jev, answers structured questions with typed output and a confidence measure attached. The account of its launch says developers have to check those scores against real outcomes on their own data first.
Reality
- Evidence26
- Adoption7
- Hype gap+48
- Incentives80
- Confidence57