DeepSeek API Price Increase: Up to 12x, Peak Hours (2026)
DeepSeek's new API prices went live 2026-08-16 16:00 UTC. Cache hits jumped 12x on V4 Pro, output 4.7x on Flash. Peak hours, new rates, and the bill math.
DeepSeek Harness (dsh): Version, Updates, Stability (2026)
DeepSeek Harness (dsh) is still on 0.1.0-rc.6, its launch build. Zero commits since 2026-08-13, no releases, no tags, issues off. Pin before you build.
Codex "Couldn't Load Its Resources": 5 Causes to Fix (2026)
Codex extension won't start? The 26.803.41515 regression broke 5 different ways. Fixed in 26.810.41047, but the error message still hides two of them.
Run Qwen 3.8 27B Locally (2026): 17 GB Total, Not 17 GB VRAM
The 17 GB figure for Qwen 3.8 27B is RAM plus VRAM, not a 16 GB card. Real GGUF sizes: 9.01 GB at 2-bit, 17.11 GB at 4-bit, measured at 7.11 tok/s.
DeepSeek Harness (dsh) Setup: Run Any Model in It (2026)
DeepSeek Harness runs as a web app, not a TUI. Install it, point it at any OpenAI-compatible gateway, and know what breaks first. Tested 2026-08-14.
DeepSeek V4 Pro 0813: Price, Open Weights, Benchmarks (2026)
V4 Pro 0813 scores 53 on Artificial Analysis at $0.06 per task, 5x under GLM-5.2. MIT weights are now public. Peak/off-peak pricing starts 2026-08-16.
GLM 5.3: Benchmarks, API Access, and Weights ETA (2026)
GLM 5.3 runs on the Coding Plan today: 1M context, 128K output, Terminal Bench 3.0 up 4.6 to 28.3, weights in ~2 weeks, thinking disabled removed.
How to Use Seedance 2.5 (2026): Prompts, Timestamps, Cost
Seedance 2.5 takes 30-second clips, 50 reference assets, and integer-second timestamps. The official spec, three real API runs, and what each clip billed.
Seedance 2.5 vs 2.0: What Changed, and When 2.0 Still Wins
Seedance 2.5 doubles clips to 30s and takes 50 references, but caps at 720p until 2026-08-17 and costs 60% more per second. Same-prompt frames from both, and real bills.
Claude Opus 5 Alternatives: Pin 4.6 or Fix Your Prompt
Opus 5 says it's done when it isn't? Opus 4.5-4.8 are still live at the same $5/$25. Symptom-to-fix table, retirement dates, a measured 1.30x token gap.
What Is a Context Window? Token Limits by Model (2026)
A context window is the token budget for one request, input plus output. One identical document measured across 9 models: 614 to 957 tokens, a 1.56x spread.