Invite-only cult of developers making futuristic remote work culture possible, today. Founded in 2016. Apply at invidelabs.com/developer.htm…
100% Remote
- Tweets 2,637
- Following 34
- Followers 500
- Likes 577
The hackathon is in full force!
Tune to this Discord server for more updates discord.com/invite/invide-re…
27
Only few hours left to AI Agent Buildathon.
100+ approved participants.
11 AM, Sun, Sep 27.
Can't wait to see the AI Agents builders can ship in a day.
luma.com/nwtepy71
#developer #event #hackathon #buildinpublic
20
GLM-5.3 costs ~9× Flash per token and scores 42% vs 33% on Artificial Analysis's Terminal-Bench 4.0. Kimi K3 costs about the same per index task. DeepSeek V4 Pro costs a third as much.
Our analysis covers benchmarks, migration and hardware.
blog.invidelabs.com/glm-5-3/
1
39
Opus 5.5 at default effort scored 51 vs GPT-6 Sol’s 48 at max on Artificial Analysis’s Intelligence Index. Cost per task: $1.34 vs $1.06. Our comparison also covers Luna’s savings and Grok 4.7’s higher output-token use.
blog.invidelabs.com/opus-5-5…
61
Jev put System One models in the spotlight. Laya, Nimble, Kev and OpenThai followed, but they differ in design, limits and test evidence.
This explainer covers what these models return and how to compare them with your current LLM or classifier.
blog.invidelabs.com/system-o…
33
GLM-5.3-Flash costs up to 50x less than Opus per output token. But does that make finished work cheaper?
We looked at API pricing, independent tests, a production rollback, and the hardware needed to self-host it.
blog.invidelabs.com/glm-5-3-…
19
DeepSeek V4.1 Flash activates 8B parameters per input token, supports a 1M context and starts at $0.003/M for cached input.
DeepSeek reports strong coding-agent results, but Terminal-Bench 4.0 puts it at 31.2 vs Opus 5.0’s 51.8.
Specs, pricing and API:
blog.invidelabs.com/deepseek…
46
Invide :: Remote Developers Community retweeted
We had an incredible time today at DevRel Pe Charcha with our DevRelFolks , conversation flow was so interesting from just opinionated “What is DevRel” 😁 to “What the Dynamics of this position is something so unique”.
Kudos to Pradeep for Hosting Us 🤝
2
2
18
540
Shopify is leaving React Native after coding agents changed its cost model.
Shop was rebuilt natively in 12 weeks, but the migration still relied on native specialists, parity checks, and extensive agent infrastructure.
More details: blog.invidelabs.com/shopify-…
37
OpenAI says its AI solved Navier-Stokes, a Millennium Prize problem still listed as unsolved.
Researchers say unpublished work went into Codex. OpenAI denies accessing specific user data, but can't rule out de-identified usage data improving its models.
blog.invidelabs.com/openai-n…
32
Jellyfin 12.0 is stable, but its database migration has no simple downgrade path.
Before upgrading, verify your backup, platform support, usernames, and plugins. Also, allow time for the required full library scan.
Read the upgrade checklist: blog.invidelabs.com/jellyfin…
26
Chrome fixed exploited V8 flaw CVE-2026-85046. CISA’s deadline is September 18.
Playwright caches, CI images, PDF services, and kiosks may still carry vulnerable browser binaries. Audit every runtime, not only desktop Chrome.
blog.invidelabs.com/chrome-c…
67
GPT-6 Astra is OpenAI’s first Critical-rated cyber model.
Its API can stop monitored agent jobs with a 403 and no general resume path. Chat Completions sits outside this monitoring system.
More details: blog.invidelabs.com/gpt-6-as…
2
1
64
Claude Fable 5.1 cuts cache reads from $1 to $0.25 per million tokens.
The migration also brings three breaking changes, requires 30-day data retention, and drops Priority Tier support.
What agent teams need to test: blog.invidelabs.com/claude-f…
43
The EU has classified ChatGPT as a Very Large Online Search Engine (VLOSE) because it searches the web.
OpenAI reported 159.1 million monthly EU users, more than triple the DSA threshold. The full classification decision remains unpublished.
blog.invidelabs.com/chatgpt-…
20
htmx 4.0 is stable, but npm still installs 2.x by default.
The cautious rollout avoids surprise upgrades for unversioned CDN users. Teams pinning 4.0 should check error-response swaps, hx-disable’s new meaning, and the 60-second timeout.
blog.invidelabs.com/htmx-4-n…
31
NVIDIA reportedly agreed to buy Hugging Face for $12.9B. Neither company has confirmed it.
Both signed July’s open-weights letter, but questions remain about private repos, mirroring, exports and accelerator neutrality.
blog.invidelabs.com/nvidia-h…
19
For local AI, Apple’s M5 Ultra Mac Studio keeps the 512GB ceiling but raises bandwidth to 1.2TB/s.
The 128GB M5 Max is $5,399, just $100 below base Ultra. The 512GB option requires the full chip; price remains unknown. for local AI
blog.invidelabs.com/m5-ultra…
30
JPEG XL is converging: Safari supports it, Firefox targets version 157, and Chromium has approved its intent.
But Chrome has no milestone, Firefox renders HDR as SDR, and lossless WebP remains close. Keep the fallbacks.
blog.invidelabs.com/firefox-…
2
5
2,128
MCP’s new roadmap names the right gaps: workload identity, progressive discovery and server-pushed events.
Some foundations shipped in July. The fixes remain proposals, and two of the Working Groups responsible for them are still forming.
blog.invidelabs.com/mcp-road…
3
140