Agentic Coding Weekly

Subscribe to get a free weekly email covering agentic coding tool and model updates. Join 500+ agentic engineers.

[Prashant Anand @primaprashant]

Added a couple of quick utilities to my terminal which uses local speech-to-text with Pi coding agent to solve the problem of writing complex command like

find largest files in the current directory:

`find . -maxdepth 1 -type f -exec du -h {} + | sort -rh | head -n 5`

[Agentic Coding Weekly @Agentic_Coding]

API pricing is $1.40 per million input tokens and $4.40 per million output

[Agentic Coding Weekly @Agentic_Coding]

Zai has released GLM-5.2

extends the context window from 200K to 1M tokens and adds High/Max thinking-effort controls

scores 81.0% on Terminal-Bench 2.1 and 62.1% on SWE-bench Pro

[Agentic Coding Weekly @Agentic_Coding]

- On the ultra-long SWE-Marathon it trails Opus 4.8 by 13%

It's the top open model on all three

[Agentic Coding Weekly @Agentic_Coding]

Issue 24 is out, covering agentic coding updates from last week:

Claude Fable 5 & Mythos 5 access pulled by US government, Kimi K2.7 Code, GLM-5.2, MiMo V2.5 Pro UltraSpeed, and MiMo code

Read it in your inbox!

[Agentic Coding Weekly @Agentic_Coding]

All Mythos-class traffic carries a mandatory 30-day data retention policy, and that data is not used for training

[Agentic Coding Weekly @Agentic_Coding]

Fable 5 is on the API and consumption-based Enterprise plans now, and included on Pro, Max, Team, and seat-based Enterprise plans only through June 22

After that it requires usage credits until capacity allows it to return as a standard plan feature.

[Agentic Coding Weekly @Agentic_Coding]

On Terminal-Bench 2.1 it scores 88.0% against Opus 4.8's 82.7%, GPT-5.5's 83.4% (Codex CLI), and Gemini 3.1 Pro's 70.7% (Gemini CLI)

[Agentic Coding Weekly @Agentic_Coding]

Anthropic has released Claude Fable 5, a new "Mythos-class" model that sits above the Opus tier

On agentic coding it scores 80.3% on SWE-bench Pro, ahead of the Mythos Preview model (77.8%), Opus 4.8 (69.2%), GPT-5.5 (58.6%), and Gemini 3.1 Pro (54.2%)

[Prashant Anand @primaprashant]

Issue 23 is out, covering agentic coding updates from last week:

MiniMax M3, Nemotron 3 Ultra, Qwen 3.7 Plus, Open Code Review, GitHub Copilot Desktop, and Uber's $1,500/month AI limit

Read it in your inbox!

[Prashant Anand @primaprashant]

DeepSWE Benchmark: Let's look into what it is first and then where it falls short

This is a benchmark for:

- longer-horizon tasks compared to SWE-Bench Pro

- tasks have a higher diversity and are contamination-free

[Prashant Anand @primaprashant]

This Month in Agentic Coding: May 2026

1. Tokens are getting both more expensive and cheaper.

The frontier proprietary models got more expensive. Gemini 3.5 Flash is 3x its predecessor 🧵

[Prashant Anand @primaprashant]

Launching paid tier for my agentic coding newsletter ($120/year)

You'll get a monthly email on the 1st of every month distilling the previous month of agentic coding updates into a single email

Read the first issue: https://t.co/UU73WHDXsn

[Agentic Coding Weekly @Agentic_Coding]

API pricing is tiered by input length:

- at ≤512K input tokens it runs $0.60/M input, $2.40/M output, and $0.12/M cached reads

- currently discounted 50% for 7 days to $0.30/$1.20/$0.06

- above 512K it doubles to $1.20/$4.80/$0.24

[Agentic Coding Weekly @Agentic_Coding]

MiniMax has released M3 with 1M context window

Headline architectural change: MSA (MiniMax Sparse Attention)

Cuts per-token compute at 1M context to 1/20 of the previous gen, yielding 9× faster prefill and 15× faster decode

[Agentic Coding Weekly @Agentic_Coding]

Unlike most models, which stalled within 30 submissions, its best result came on submission 145. It also ran ~12 hours to reproduce an ICLR paper