Vienna-based nonprofit fostering interdisciplinary innovation across science, technology and entrepreneurship. Home of the WAISS.
Hackathons
OpenClaw HACK_001
HACK_002
Peter Wimberger
@topsrek
Projects
Intercortex, Machine Literacy
LinkedIn
/in/pwimberger
[QR]
WLAN
[Network]
[Password]
[Password]
WAISS · #102 / 28
Agenda
Tonight
01AI recap of the month20 min
02Show & Tell · short talks[xx min]
03Open round & questionsopen end
WAISS · #103 / 28
Website
waiss.teloscircle.com
Want to present something next time?
Open a pull request on github.com/topsrek/waissor email events@teloscircle.com
HTML preferred · PPT and PDF work too
WAISS · #104 / 28
Releases · Meta · 09 Nov
Muse Ember
Artificial Analysis@ArtificialAnlys
Meta’s newest model Muse Ember scores 36 on the Artificial Analysis Intelligence Index, matching Claude Haiku 5.5 (max) at a third of the Cost per Task
19:07 · 9 Nov 2026
WAISS · #105 / 28
Releases · Meta
Muse Ember
Haiku-level, at a third of the cost
Intelligence IndexArtificial Analysis, max effort
Muse Ember
36
Claude Haiku 5.5
36
DeepSeek V4.1 Flash
33
GLM-5.3-Flash
28
Cost per taskUSD per index task, lower is better
Muse Ember
$0.14
Claude Haiku 5.5
$0.41
DeepSeek V4.1 Flash
$0.60
GLM-5.3-Flash
$0.58
Coding Agent IndexIn Muse Code / Claude Code / Codex
Muse Ember
46
Claude Haiku 5.5
37
GPT-6 Luna
31
$0.10 / $0.50 per 1M tokens in / out. Weak spot: knowledge (AA-Omniscience 21% vs. 35% for Haiku), it declines 64% of questions but hallucinates half as often. Source: x.com/ArtificialAnlys/status/2119854348587123007
WAISS · #106 / 28
Releases · Anthropic
Fable 5.2
Claude@claudeai
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
19:57 · 9 Nov 2026
WAISS · #107 / 28
Releases · Anthropic
Fable 5.2
Ahead almost everywhere, 20% cheaper than 5.1
FrontierSWE v2Agentic coding
Fable 5.2
69.9
GPT-6 Astra
65.5
Opus 5.5
62.3
Fable 5.1
56.3
Humanity’s Last ExamExpert questions, with tools
Fable 5.2
74.2
Opus 5.5
67.7
Fable 5.1
65.6
GPT-6 Astra
57.2
RoboDojo SimRobot control in simulation
Fable 5.2
64.9
Opus 5.5
39.4
GPT-6 Astra
29.8
Fable 5.1
27.6
Score in %, vendor-reported, highest effort per model. Gemini 4 Carbon leads AutomationBench (53.9 vs 51.4). Source: anthropic.com/claude-fable-and-mythos-5-2
WAISS · #108 / 28
Benchmarks · Artificial Analysis
Fable 5.2
New #1 on the Intelligence Index
Claude Fable 5.2
65
Gemini 4 Carbon
60
Claude Opus 5.5
59
GPT-6 Astra
57
GPT-6.1 Sol
55
Claude Sonnet 5.5
54
Muse Wonder Full
50
Claude Fable 5.1
49
Index v5.0.1, measured independently, highest effort per model. Fable 5.1 → 5.2: +15 points in ten weeks. Source: artificialanalysis.ai, 10 Nov 2026
WAISS · #109 / 28
Benchmarks · Epoch AI
Fable 5.2
Biggest jump of the year on Epoch’s Capabilities Index
Every model released since 10 Nov 2025; ink = new record on release, line = best model so far. Fable 5.2: 174 (90% CI 170–179), +5 over GPT-6 Astra. Source: epoch.ai/eci, eci_scores.csv of 10 Nov 2026
WAISS · #110 / 28
Prices · USD per 1M tokens
Fable 5.2
2× Opus per token, cheaper per task
MODEL
LAB
INPUT
OUTPUT
COST / TASK*
Fable 5.2
Anthropic
$8.00
$40.00
$17.94
Opus 5.5
Anthropic
$4.00
$20.00
$24.77
Sonnet 5.5
Anthropic
$2.00
$10.00
$16.18
Haiku 5.5
Anthropic
$0.10
$0.50
$3.18
GPT-6 Astra
OpenAI
$10.00
$50.00
$21.28
GPT-6.1 Sol
OpenAI
$2.00
$10.00
$3.61
GPT-6.1 Terra
OpenAI
$1.00
$5.00
$2.94
DeepSeek V4.1 Flash
DeepSeek
$0.30
$1.20
$0.66
Muse Ember
Meta
$0.10
$0.50
$0.15
List prices, standard tier, from the providers’ pages via artificialanalysis.ai, 10 Nov 2026. *Average cost of one Vals Index task (vals.ai, 9 Nov 2026): fewer tokens can beat a lower price.
WAISS · #111 / 28
Benchmarks · Vals AI
Fable 5.2
#1 on the GDP-weighted Vals Index, at 72% of Opus’s cost
Accuracy on agentic finance, coding, legal and tax tasks, weighted by each sector’s share of U.S. GDP; line = Pareto frontier, bars = ±1 standard error. Source: vals.ai/benchmarks/vals_index, v2.2 of 9 Nov 2026
WAISS · #112 / 28
Benchmarks · Vals AI
Fable 5.2
Same accuracy as Opus 5.5, in half the time
MODEL
ACCURACY
COST / TASK
TOKENS / TASK
TIME / TASK
Fable 5.2
71.3%
$17.94
31.3M
43m 42s
Sonnet 5.5
70.9%
$16.18
78.6M
1h 26m
Gemini 4 Carbon
70.5%
$10.51
18.3M
47m 45s
Opus 5.5
70.1%
$24.77
59.5M
1h 25m
Gemini 4 Argon
69.4%
$13.37
23.3M
44m 58s
Muse Wonder Full
66.8%
$6.81
18.4M
59m 19s
GPT-6 Astra
65.0%
$21.28
8.1M
33m 39s
Vals Index leaderboard, top 7. The top four are within one standard error (±1.0 points) of each other. Source: vals.ai/benchmarks/vals_index, v2.2 of 9 Nov 2026
WAISS · #113 / 28
Community
Fable 5.2
4.7× less compute than Opus 5.5
Dominik Scholz@dom_scholz
Opus 5.5 vs. Fable 5.2
waaaay quicker, almost 0-shot from fable
Fabulous
waaaay quicker, almost 0-shot from fable
Fabulous
00:20 · 10 Nov 2026
WAISS · #114 / 28
Opus 5.5
“For most coding and knowledge tasks, we recommend starting with Opus 5.5, our strongest daily driver.”
— @claudeai, 9 Nov 2026
WAISS · #115 / 28
Releases
[03 Oct]
[Headline of the story]
[One sentence of context]
WAISS · #116 / 28
Drama
[@]
[Name][@handle]
[The tweet, in full. Short enough to read from the back row.]
[10:42 · 3 Oct 2026]
WAISS · #117 / 28
Drama
[@]
[Name][@handle]
[First tweet]
[10:42 · 3 Oct 2026]
[@]
[Name][@reply]
[The reply, or the other side of it]
[10:42 · 3 Oct 2026]
WAISS · #118 / 28
News
[@]
[Name][@handle]
[Tweet text with a screenshot or chart attached.]
[Image / screenshot]
[10:42 · 3 Oct 2026]
WAISS · #119 / 28
Roundup
[What changed in everyday work]
Releases[Item A][One sentence on why it matters]
News[Item B][One sentence on why it matters]
Drama[Item C][One sentence on why it matters]
WAISS · #120 / 28
Number of the month
[73 %]
[What the number means, in one sentence]
Source: [link]
WAISS · #121 / 28
Show & Tell
[Talk title]
[Name]
[Role · @handle]
[Screenshot / demo]
WAISS · #122 / 28
[One sentence that sticks.]
— [Source]
WAISS · #123 / 28
Open round
What surprised you this month?
WAISS · #124 / 28
Discussion
Are the frontier labs pacing themselves?
Every model in Epoch’s list, 10 Nov 2025 – 10 Nov 2026; filled = new best of that lab. Since April, Anthropic has set a new lab best every 2–7 weeks. Source: epoch.ai/eci
WAISS · #125 / 28
Discussion
Will OpenAI & Anthropic
still dominate in a year?
Best Epoch Capabilities Index score per lab. A year ago: Google −6, Meta −19. Not in Epoch yet: Gemini 4 Carbon, 0.9 points behind Fable 5.2 on the Vals Index. Source: epoch.ai/eci
WAISS · #126 / 28
@SadAlbert10 on X
See you next time.
Telos Circle · Links
links.teloscircle.com
Discord
discord.gg/vjTeFrPsN
Slides · GitHub
github.com/topsrek/waiss
Want to present next time? Talk to Peter.
WAISS · #128 / 28