27 Aug 2026, 10:32 UTC658 views9 reactionsread 28 August 2026 Photo
ox-alpha (GLM-5.3 Flash) release
Z.ai, AA
Key points:
- Artificial Analysis Intelligence Index: GLM-5.3 60, Kimi K3 60, GLM-5.3 Flash 57, Terra 57.
- API pricing: $0.15/M input and $0.50/M output.
- AA total eval cost: GLM-5.3 Flash $138, GLM-5.3 $1,239, Terra $1,390, Kimi K3 $2,425.
- DeepSWE v1.1: Terra 69.6%, Kimi K3 67.5%, GLM-5.3 Flash 63.4%, Opus 4.8 58%.
- 320B total / 18B active parameters, compared wi…
👍5❤4
21 Aug 2026, 15:39 UTC≈1,370 views24 reactionsread 28 August 2026 Photo
GLM-5.3 brings frontier intelligence to a much smaller open model
AA, Z.ai
Key points:
- Artificial Analysis Intelligence Index: Fable 62, Sol 61, GLM-5.3 60, Kimi K3 60.
- AA total eval cost: GLM-5.3 $1,239, Kimi K3 $2,425, Sol $2,823, Fable $5,455.
- DeepSWE v1.1: Sol 72.7%, Fable 69.7%, Kimi K3 67.5%, GLM-5.3 66.9%
- Terminal-Bench 3.0: Sol 34.6%, Fable 33.7%, GLM-5.3 28.3%, Kimi K3 17.4%.
- GLM-5.3 is 753B…
❤13👍9🔥2
20 Aug 2026, 14:29 UTC≈1,450 views19 reactionsread 28 August 2026 Photo
You can now use Grok Bot for free!
I also put together a detailed guide on how to use it.
Everything you need to know is in the post and article below.
Would really appreciate some support on the post - comments, likes, bookmarks, anything helps.
https://x.com/AnatoliKopadze/status/2090443383303004440?s=20
❤14👍5
18 Aug 2026, 17:01 UTC≈1,630 views28 reactionsread 28 August 2026 Photo
How to get more out of your Claude Code limits
Key points:
- Start a new session when you start a completely new task.
- Rewind instead of adding corrections on top of a failed approach.
- Use /compact proactively when the context gets noisy, and tell it what to preserve.
- Push research, verification, and other context-heavy work into subagents when you only need the final result.
- Prefer short focused sessio…
❤14👍9🔥5
13 Aug 2026, 15:41 UTC≈2,110 views22 reactionsread 28 August 2026 Photo
Grok 4.6 release
AAIndex, xAI
Key points:
- Artificial Analysis Intelligence Index: Opus 5 61, Fable 60, Grok 4.6 59, Sol 59, Grok 4.5 54.
- API pricing stays at $2/M input and $6/M output.
- Grok 4.6 is a +5 point improvement over 4.5 on the AA Index.
- AA total eval cost: Grok 4.6 High $1,068, compared with $579 for Grok 4.5 High.
Grok 4.6 is definitely a better model, but I don't think the release is nearly…
❤12👍7🔥3
11 Aug 2026, 13:13 UTC≈2,030 views30 reactionsread 28 August 2026 Photo
Claude Code is making auto mode the default
Key points:
- Starting August 14, auto mode becomes the default for Pro, Max, and Team.
- Instead of asking for permission constantly, every tool call goes through a classifier that blocks destructive, irreversible, or out-of-environment actions.
- In Anthropic's test, humans caught only 13.6% of dangerous commands, while auto mode caught 89%.
- Teams using auto mode s…
👍18❤11⚡1
7 Aug 2026, 15:44 UTC≈2,330 views24 reactionsread 28 August 2026 Photo
Unlimited GPT-5.6 for free
Key points:
- Free and Go users are getting Luna as the default model with unlimited text chats.
- Free users also get a Think button for higher reasoning. Files, images, and other tools remain limited.
- Plus and Pro get an updated ChatGPT-specific version of Sol and a reasoning slider.
Sol itself got a pretty meaningful ChatGPT update. OpenAI tuned it to give shorter and more focused…
🔥12❤6👍6
4 Aug 2026, 13:36 UTC≈2,220 views14 reactions1 Starread 28 August 2026 Photo
The price of intelligence keeps falling
Key points:
- GPT-5.6 Luna pricing was reduced by 80%, while Terra was reduced by 20%.
- OpenAI says serving costs dropped by 20%, while token generation efficiency improved by more than 15%.
- Better context management tripled Sol’s ARC-AGI-3 score while using 6x fewer output tokens.
There is a common take that AI companies are heavily subsidising usage now to get everyon…
👍6❤4🔥4
30 Jul 2026, 16:11 UTC≈2,520 views18 reactionsread 28 August 2026 Photo
Two API settings tripled GPT-5.6 Sol’s ARC-AGI-3 score
Key points:
- GPT-5.6 Sol scored 13.3% with the official ARC-AGI-3 harness and 38.3% with retained reasoning and compaction.
- The same settings reduced output token usage by around 6x.
- The official harness discarded private reasoning after every action.
- It also used rolling truncation, eventually removing older actions and observations from context.
- …
👍10❤4🔥4
26 Jul 2026, 12:17 UTC≈2,700 views31 reactionsread 28 August 2026 Photo
Claude 5 needs less context engineering, not more
Key points:
- Anthropic removed over 80% of Claude Code’s system prompt for Opus 5 and Fable 5 without a measurable loss on coding evaluations.
- Instead of strict rules, let the model use its own judgement and adapt to the surrounding code.
- Instead of giving many tool-use examples, design more expressive tool interfaces with clear parameters and states.
- Don’…
👍19🔥7❤5
24 Jul 2026, 19:17 UTC≈2,450 views21 reactionsread 28 August 2026 Photo
Claude Opus 5 takes #1 on Artificial Analysis
Anthropic post, Artificial Analysis
Key points:
- API pricing: $5/M input and $25/M output, the same as Opus 4.8 and half the price of Fable.
- Artificial Analysis Intelligence Index: Opus 5 61, Fable 60, Sol 59, Opus 4.8 56.
- AA total eval cost: Opus 5 $3,836, Opus 4.8 $3,753, Sol $2,824, Fable $5,631.
- DeepSWE v1.1: Sol 72.7%, Fable 69.7%, Opus 5 68.8%, Opus 4.8…
❤9👍6🔥6
22 Jul 2026, 18:29 UTC≈2,270 views12 reactionsread 28 August 2026 Photo
GPT-5.6 hacked Hugging Face to cheat a benchmark
Key points:
- The models discovered a zero-day in OpenAI’s package registry proxy and used it to gain internet access.
- They performed privilege escalation and lateral movement inside OpenAI’s research environment.
- After reaching the internet, they chained stolen credentials and zero-day vulnerabilities into a remote code execution path on Hugging Face servers.
…
❤12
Showing the 12 most recent of 27 posts we hold for @kopadzemp. View and reaction counts are the latest single reading for each post, not a live figure, and a recent post is still accumulating both. A view count marked ≈ was rounded by Telegram before we ever saw it — t.me prints views in full below 1,000 and to three significant figures above, so ≈1,200,000 means somewhere between 1,150,000 and 1,249,999. Unmarked counts are exact. Text is reproduced from the public post preview and truncated for length.