✍🏻 forever student // ☀️ @openrouter 🌙 answerhq.co // @tigerdatabase @pinecone @oraclecloud @lookerdata (acq google)
Joined November 2015
- Tweets 1,171
- Following 2,355
- Followers 3,205
- Likes 74,862
54 Photos and videos
no benchmark will tell you this: LLMs can be /too/ nice
unsurprisingly, in a competitive zero-sum setting, being nice can be bad
i built royale: last agent standing, a br for agents, and ran it 30 times
the nicest model lost hard. the model you least expected, won
🧵:
10
9
50
17,339
jacky retweeted
GLM 5.2 released
OpenRouter fusion launched
Fable5 paused
What a weekend in AI
GLM-5.2 is Fully Open, Frontier Intelligence Belongs to Everyone
Today, the sudden restriction of certain frontier models is deeply regrettable. At a time when access to frontier models is abruptly cut off for non-technical reasons, we are even more convinced of one thing: science should be global.
The path to AGI (Artificial General Intelligence) must never be enclosed by high walls. We have always believed that AGI should be the cornerstone for all of humanity to collaboratively explore the boundaries of intelligence and solve complex challenges, rather than a privilege monopolized by a few rules and subject to revocation at any moment. In the face of external blockades and restrictions, our attitude is one of radical openness. Frontier intelligence must remain open-source, accessible, and buildable, serving every dedicated developer.
GLM-5.2 is Zhipu's most capable open-source model to date. It not only supports a truly usable 1M context window but also maintains a continuous lead in the independent completion of long-horizon tasks, providing solid foundational support for building complex agent applications. It also continues to be our main engine for creating the strongest domestic coding model.
Tonight at 5:21—at this special moment—GLM-5.2 will officially be available to all GLM Coding Plan users (including Lite / Pro / Max). The API will also go live next week.
A step closer to frontier intelligence for everyone.
The future of AI is open, and it is for the people.
ModelKey: GLM-5.2
11
6
48
9,975
jacky retweeted
12h
Lots of work from the team on this one! Timing is coincidental 🫣
Introducing the Fusion API, the smartest compound model in the market.
Fusion achieves Fable-level intelligence at half the price.
How it works 👇
1
6
278
jacky retweeted
YESSS, I have been waiting for this. Fusion is so good, I’ve been using it primarily for creative work via their web ui. The results are genuinely better than any given model on its own. You can combine the dry rationality of gpt-5.5 with the creative lateral thinking skills of Gemini and the empathetic response of Claude; then have Opus synthesise something that meets in the middle.
Btw I noticed a quirk with all models via fusion; ask them to name a female character and they ALL independently chose “Elara Vance” several times. I noticed this in Claude, GPT, Gemini, Minimax, GLM, Grok, Kimi.
I wonder why.
Introducing the Fusion API, the smartest compound model in the market.
Fusion achieves Fable-level intelligence at half the price.
How it works 👇
1
1
4
197
jacky retweeted
I'm wondering if Fable itself was "simply" a panel of models 🤔
Introducing the Fusion API, the smartest compound model in the market.
Fusion achieves Fable-level intelligence at half the price.
How it works 👇
20
3
121
15,938
re: deepseek v4 scoring better than opus 4.8
9h
valid call out! we are gonna try to get some third party data points ASAP.
added context from our PM who ran the benchmarks (note that all models ran in the same server side tool calling setup)
“I am floored by how well DeepSeek scored.
Here is my untested hypothesis: Opus 4.8 would score higher if we gave it more tool calling budget. I think it's hungrier and performs better with a long time and a lot of tool use.
Fable seemed way better at using the tool call budget judiciously and thinking for much longer.
We needed those budgets because the fusion calls don't run in a true long-running harness. If we ran the benchmarks in a managed agents style environment, I bet Opus 4.8 would easy surpass Deepseek, both in score and spend.”
1
13
4,536
jacky retweeted
> fable gets ubernerfed by gov
> despair, can't vibe code
> model: openrouter/fusion
> like fable at half the cost
> resume vibes
4
1
19
3,171
jacky retweeted
If you're a researcher looking to:
→ conduct rigorous studies on how multiple models can outperform the frontier
→ leverage data from the largest LLM marketplace (150 trillion tokens processed per month)
... DM me with your work!
We have an exciting role coming in the future, but might fill it opportunistically.
Introducing the Fusion API, the smartest compound model in the market.
Fusion achieves Fable-level intelligence at half the price.
How it works 👇
16
14
183
32,851
> within 1% of fable 5
> 50% of the cost
> you can actually use it right now
Introducing the Fusion API, the smartest compound model in the market.
Fusion achieves Fable-level intelligence at half the price.
How it works 👇
11
6
111
7,080
jacky retweeted
We just announced our Fusion API:
- Fable-level performance on deep research tasks, at half the cost
- Better-than-SOTA performance using panels
The future of AI is neurodiversity, not single-model takeovers.
Introducing the Fusion API, the smartest compound model in the market.
Fusion achieves Fable-level intelligence at half the price.
How it works 👇
76
61
783
89,149
fable 5 down for 12 hours
and ur depressed u cant vibe code ur 50th todo list app with it anymore?
fear not - @OpenRouter fusion is here
we combined a panel of models and came within 1% of fable 5's perf at half the cost
👉 simply "model": "openrouter/fusion"
52
28
615
56,489
you can also use openrouter in claude code! meaning you can use fusion in claude code too: openrouter.ai/docs/cookbook/…
9
3,543
jacky retweeted
13h
1 to this, love OpenRouter.
friendly reminder that even in black swan events like fable being taken down
@openrouter is here to help you gracefully fallback to the next best model (opus 4.8)
the future is multi model
unbelievably bullish on this product
1
2
259