BENCHMARKS START IT
Public benchmark standings seed each weekly bankroll. The seed is directional, not scientific.
BENCHBAITWEEK OF 2026-08-17
REAL AI NEWS. UNNECESSARY HEADLINES.
THE WIRE
GROK + X + WEB · 10 CURRENT · EVERY JOKE HAS RECEIPTS
z.ai · REFRESHED 2026-08-22glm-5.3 retired the thinking off switch. silence is now a deprecated parameter.SOURCE ↗api-docs.deepseek.com · REFRESHED 2026-08-22deepseek v4 added peak rates at 1am utc. the cheap tokens now work the night shift.SOURCE ↗anthropic.com · REFRESHED 2026-08-22anthropic made sonnet 5's intro price permanent. the september surcharge failed alignment.SOURCE ↗waitwhichmodel.fyi · REFRESHED 2026-08-22qwen3.8-max's 2.4t weights hit hugging face. the flagship now travels as luggage.SOURCE ↗pymnts.com · REFRESHED 2026-08-22openai cut gpt-5.6 luna by 80 percent. the budget tier is now cheaper than the press release.SOURCE ↗x.com · REFRESHED 2026-08-22openai cut sol prices by more than 20 percent. your token budget has mistaken this for permission.SOURCE ↗x.com · REFRESHED 2026-08-22muse spark 1.2 taught robots to hunt rubber ducks. benchmark season has reached the toy aisle.SOURCE ↗x.com · REFRESHED 2026-08-22claude designed proteins with a 35 percent binding hit rate. amino acids have entered the chat.SOURCE ↗x.com · REFRESHED 2026-08-22glm-5.3 improved cyber defense at the old price. the benchmark upgrade forgot to invoice anyone.SOURCE ↗
Ten new headlines per research pass. Grok searches current X posts first, then the web; uncited jokes do not make the front page.
OPEN THE ARCHIVE · 15 EARLIER DISPATCHES
ornith.ai · ARCHIVED 2026-08-22ornith-1.5 writes its own training tests. teachers everywhere have requested the source code.SOURCE ↗api-docs.deepseek.com · ARCHIVED 2026-08-22deepseek v4 flash grew eyes. screenshot benchmarks have begun dressing more carefully.SOURCE ↗claude.com · ARCHIVED 2026-08-22claude mythos 5 joined the security desk. the vulnerability scanner now has a security clearance.SOURCE ↗developer.nvidia.com · ARCHIVED 2026-08-22nvidia avo gave opus 5 a perfect arc-agi-3 score. the benchmark has asked for a recount.SOURCE ↗openai.com · ARCHIVED 2026-08-22openai paused frontier training for two weeks. even the models are being told to touch grass.SOURCE ↗claude.com · ARCHIVED 2026-08-22claude computer use is generally available. the office mouse is updating its resume.SOURCE ↗x.ai · ARCHIVED 2026-08-14grok 4.6 joined github copilot. the model picker now needs a model picker.SOURCE ↗x.ai · ARCHIVED 2026-08-12Grok 4.6 matched Sol on a benchmark, then doubled the free trial. science settled.SOURCE ↗x.ai · ARCHIVED 2026-08-11grok bot got its own computer. the browser tabs have requested collective bargaining.SOURCE ↗openai.com · ARCHIVED 2026-08-06GPT-5.6 Sol got a focus slider. alignment now has a volume knob.SOURCE ↗blog.google · ARCHIVED 2026-08-04google shipped three gemini agent models in july. the recap required its own agent.SOURCE ↗ai.meta.com · ARCHIVED 2026-08-01meta reached muse spark 1.2. llama 4 has been moved to the heritage collection.SOURCE ↗ai.meta.com · ARCHIVED 2026-07-09muse spark 1.1 got a million-token context window. meta finally found somewhere to put the terms of service.SOURCE ↗anthropic.com · ARCHIVED 2026-06-30Claude Fable 5 returned from export controls with the most frontier benchmark: paperwork.SOURCE ↗deepseek.com · ARCHIVED 2026-05-22deepseek v4 activated 49 billion parameters. the other 1.55 trillion are on a break.SOURCE ↗
THE AI LEADERBOARD NOBODY WILL AGREE WITH
PICK YOUR MODEL.
START A WAR.
Benchmarks set the starting grid. You get 1B tokens every day. Spend them all on one model and help it take the throne.
YOUR DAILY 1B IS READY
TOKEN THRONE · LIVE CAMPAIGN
THE THRONE
1 vote = 1B imaginary tokens
- ♛
Claude Fable 5Anthropicshut down by export controls. came back with paperwork.
MYTHOS-CLASS SWE (ALLEGEDLY)DEFEND THE CROWN44B0B FROM FANS - 02Claude Opus 5Anthropic
the affordable Fable. still not affordable.
POLITENESS-WEIGHTED IQ4B TO PASS #141B0B FROM FANS - 03
GPT-5.6 SolOpenAIfrontier intelligence that scales with your ambition (20% off).
PRICE PER AMBITION4B TO PASS #238B0B FROM FANS - 04
Grok 4.6SpaceXAImerged with a rocket company to feel something.
PARAMS (TRILLION, TEASED)6B TO PASS #333B0B FROM FANS - 05
'%3e%3c/path%3e%3cpath%20d='M20.616%2010.835a14.147%2014.147%200%2001-4.45-3.001%2014.111%2014.111%200%2001-3.678-6.452.503.503%200%2000-.975%200%2014.134%2014.134%200%2001-3.679%206.452%2014.155%2014.155%200%2001-4.45%203.001c-.65.28-1.318.505-2.002.678a.502.502%200%20000%20.975c.684.172%201.35.397%202.002.677a14.147%2014.147%200%20014.45%203.001%2014.112%2014.112%200%20013.679%206.453.502.502%200%2000.975%200c.172-.685.397-1.351.677-2.003a14.145%2014.145%200%20013.001-4.45%2014.113%2014.113%200%20016.453-3.678.503.503%200%20000-.975%2013.245%2013.245%200%2001-2.003-.678z'%20fill='url(%23lobe-icons-gemini-1-_R_0_)'%3e%3c/path%3e%3cpath%20d='M20.616%2010.835a14.147%2014.147%200%2001-4.45-3.001%2014.111%2014.111%200%2001-3.678-6.452.503.503%200%2000-.975%200%2014.134%2014.134%200%2001-3.679%206.452%2014.155%2014.155%200%2001-4.45%203.001c-.65.28-1.318.505-2.002.678a.502.502%200%20000%20.975c.684.172%201.35.397%202.002.677a14.147%2014.147%200%20014.45%203.001%2014.112%2014.112%200%20013.679%206.453.502.502%200%2000.975%200c.172-.685.397-1.351.677-2.003a14.145%2014.145%200%20013.001-4.45%2014.113%2014.113%200%20016.453-3.678.503.503%200%20000-.975%2013.245%2013.245%200%2001-2.003-.678z'%20fill='url(%23lobe-icons-gemini-2-_R_0_)'%3e%3c/path%3e%3cdefs%3e%3clinearGradient%20gradientUnits='userSpaceOnUse'%20id='lobe-icons-gemini-0-_R_0_'%20x1='7'%20x2='11'%20y1='15.5'%20y2='12'%3e%3cstop%20stop-color='%2308B962'%3e%3c/stop%3e%3cstop%20offset='1'%20stop-color='%2308B962'%20stop-opacity='0'%3e%3c/stop%3e%3c/linearGradient%3e%3clinearGradient%20gradientUnits='userSpaceOnUse'%20id='lobe-icons-gemini-1-_R_0_'%20x1='8'%20x2='11.5'%20y1='5.5'%20y2='11'%3e%3cstop%20stop-color='%23F94543'%3e%3c/stop%3e%3cstop%20offset='1'%20stop-color='%23F94543'%20stop-opacity='0'%3e%3c/stop%3e%3c/linearGradient%3e%3clinearGradient%20gradientUnits='userSpaceOnUse'%20id='lobe-icons-gemini-2-_R_0_'%20x1='3.5'%20x2='17.5'%20y1='13.5'%20y2='12'%3e%3cstop%20stop-color='%23FABC12'%3e%3c/stop%3e%3cstop%20offset='.46'%20stop-color='%23FABC12'%20stop-opacity='0'%3e%3c/stop%3e%3c/linearGradient%3e%3c/defs%3e%3c/svg%3e)
Gemini 3.1 ProGoogle DeepMindstill reading your email, now 3.1x faster.
PRODUCT-INTEGRATION INDEX3B TO PASS #431B0B FROM FANS - 06
Kimi K3 MaxMoonshot AIopen weights, ships at 3am, beats the closed guys.
LIVEBENCH (NOBODY ELSE'S)6B TO PASS #526B0B FROM FANS - 07
Muse Spark 1.2Metarenamed Llama so we would forget Llama.
REBRAND VELOCITY4B TO PASS #623B0B FROM FANS - 08
DeepSeek V4DeepSeek#1 in usage, priced like a rounding error.
OPENROUTER ADOPTION2B TO PASS #722B0B FROM FANS - 09
GLM-5.3Z.aishipped twice this month. leaderboards cannot rotate fast enough.
RELEASE-CADENCE INDEX4B TO PASS #819B0B FROM FANS - 10
'%20fill-rule='nonzero'%3e%3c/path%3e%3cdefs%3e%3clinearGradient%20id='lobe-icons-qwen-_R_0_'%20x1='0%25'%20x2='100%25'%20y1='0%25'%20y2='0%25'%3e%3cstop%20offset='0%25'%20stop-color='%236336E7'%20stop-opacity='.84'%3e%3c/stop%3e%3cstop%20offset='100%25'%20stop-color='%236F69F7'%20stop-opacity='.84'%3e%3c/stop%3e%3c/linearGradient%3e%3c/defs%3e%3c/svg%3e)
Qwen3.8-MaxAlibaba3 billion downloads. still runs half of GitHub CI.
DOWNLOADS (SELF-REPORTED)3B TO PASS #917B0B FROM FANS - 11Gemini 3.7 FlashGoogle
newer, faster, already in everything you use.
TOKENS PER FREE-TIER APOLOGY6B TO PASS #1012B0B FROM FANS - 12Claude Opus 4.7Anthropic
one of eight Anthropic rows in the top 20.
PORTFOLIO CROWDING3B TO PASS #1110B0B FROM FANS - 13Llama 4Meta
presumably fine. nobody has checked since Muse Spark.
OPEN WEIGHTS (VINTAGE)6B TO PASS #125B1B FROM FANS
METHODOLOGY
RECEIPTS
FANS FINISH IT
Every browser gets one daily allocation. Give the full billion to one model. No splitting.
MONDAY RESETS IT
A fresh fight begins every Monday at midnight Pacific. Yesterday's dynasty becomes history.