Desk · Labs & reverse engineering

Research

Technical deep dives, reproducible tests, and tool evaluations.

Articles: 177
Topics: 20
Page: 3 / 23

Data graphic: ChatGPT Pro 500 $500 a month adds Ultrafast Codex, up to 8x faster tokens than Standard on GPT-6 Astra, and only Pro 500 has it, while Pro 100 $100, 5x Plus usage and Pro 200 $200, 10x from 30 Oct do not. The verdict is to pay $500 only if waiting on token generation is your bottleneck.

ChatGPT Pro 500: What $500 a Month Buys

ChatGPT Pro 500 costs $500 a month and adds Ultrafast Codex, up to 8x faster. Here is what it changes, what Pro 200 lost, and who should pay.

Sep 30 · 6 min

Data poster comparing Claude Sonnet 5.5 and Sonnet 5: $2 input, $10 output, $0.20 cached read per million tokens, 1M context; Sonnet 5.5 at medium effort beats Sonnet 5 at high for under a fifth of the cost.

stableResearch Anthropic

Claude Sonnet 5.5: Same Price as Sonnet 5, More Work per Dollar — and How to Run It Efficiently

Sonnet 5.5 keeps Sonnet 5's $2/$10 pricing, runs 30% faster and nears Opus 5.5 on benchmarks. How to set effort, cache and batch to cut cost per task.

Sep 29 · 7 min

Data graphic: 4-bit model file sizes against the memory budget for local LLMs on 16GB hardware. A 16GB Mac allows about 10.7GB of GPU memory by default, against roughly 15GB on a 16GB NVIDIA card, so Laguna XS 2.1 20.3GB and Qwen3.8-27B 16.8GB do not fit a 16GB card, while Devstral Small 2 24B 14.3GB , Gemma 4 26B A4B 14.2GB and gpt-oss-20b 12.8GB do. Practical context is 32K–64K for gpt-oss-20b, 8K–16K for Gemma 4 26B A4B and 4K–8K for Devstral Small 2 on the card, and 32K for Qwen3.5-9B ~5.6GB and 16K–32K for Gemma 4 12B 6.98GB on the Mac. The RTX 5060 Ti has 448GB/s of bandwidth to the RTX 4060 Ti's 288GB/s, and the verdict is to run gpt-oss-20b or Gemma 4 26B A4B on a 16GB card and Qwen3.5-9B on a 16GB Mac.

watchResearch Local Llm Inference

Best Local LLM for 16GB in 2026: RTX 5060 Ti, 4060 Ti and Mac mini M4 (Coding First)

gpt-oss-20b and Gemma 4 26B A4B lead on 16GB NVIDIA cards, while a 16GB Mac mini M4 is better served by Qwen3.5-9B. File sizes, context limits and VRAM math.

Sep 24 · 13 min

Data graphic: DGX Spark vs Mac Studio M5 Ultra vs Ryzen AI Max+ 395 for local LLMs. On gpt-oss-120b in the same llama.cpp test, DGX Spark prefills about 5x faster than Strix Halo 1,723 vs 340 tok/s , while decode is close 38.6 vs 34.1 tok/s ; the Mac Studio M5 Ultra has 1.2 TB/s of memory bandwidth against 273 and 256 GB/s, and price per GB of unified memory runs from about $27 Strix Halo boxes and $37 DGX Spark to $57 M5 Ultra base, 96GB and $42 M5 Ultra, 256GB .

watchResearch Nvidia Dgx Spark

DGX Spark vs Mac Studio M5 Ultra vs Ryzen AI Max+ 395: Which Box for Local LLMs?

Price per GB, memory bandwidth, prefill vs decode on gpt-oss-120b, clustering and software stacks: which unified-memory desktop fits your local LLM work.

Sep 24 · 14 min

Data graphic: Ollama Cloud's $20 Pro plan includes $60 of token credits 3x the price , and 60 Kimi K3 sessions a month cost $100 on Ollama Max versus about $252 on OpenCode Go and about $271 on OpenRouter. A table of 20 sessions a month shows OpenRouter cheapest for DeepSeek V4.1 Flash $2 , OpenCode Go for MiniMax M3 $10 , and Ollama Pro for GLM-5.3 $20 and Kimi K3 $46 ; plans are Ollama Pro $60 credits for $20, Max $300 for $100, OpenRouter's 5.5% top-up fee, and OpenCode Go's $15 to $60 per-model caps for $10.

watchResearch Ollama Cloud

Ollama Cloud vs OpenRouter vs OpenCode Go: Cheapest Way to Run Open Models for Coding

We priced the same coding-agent session on Ollama Cloud, OpenRouter and OpenCode Go. Ollama's new credit plans win on flagships; Go wins on cheap models.

Sep 24 · 15 min

Data graphic: GitHub Copilot Pro $10/mo, 1,500 AI credits worth $15 vs Pro+ $39/mo, 7,000 credits worth $70, 3,100 of them flex and Max $100/mo, 20,000 credits , at $0.01 per credit, with a bar chart of estimated credits per 25-turn medium agent task ranging from about 5 on GPT-6 Luna to about 102 on Claude Sonnet 5, 180 on Opus 5.5, 419 on Fable 5.1 and 512 on GPT-6 Astra. A table shows tasks per month per plan: Pro+ buys about 68 Sonnet 5, 38 Opus 5.5, 16 Fable 5.1 or 13 GPT-6 Astra tasks, while Pro cannot use Opus, Fable or GPT-6 Astra at all.

watchResearch Github Copilot

GitHub Copilot Pro vs Pro+ After AI Credits: What $10 vs $39 Actually Buys

Copilot now bills in AI credits at $0.01 each. Here is what Pro's 1,500 and Pro+'s 7,000 credits buy per model, and when Pro+ beats Claude or Cursor.

Sep 24 · 15 min

watchResearch Claude Code

Claude Code vs Codex CLI vs Antigravity CLI vs OpenCode: Coding Agents Compared (2026)

Gemini CLI is gone for consumer plans. We compare Claude Code, Codex CLI, Google's agy and OpenCode on price, models, license, sandboxing and MCP.

Sep 24 · 15 min

Data graphic: ChatGPT Pro's $200 plan drops to 10x Plus usage from 30 October, while Google AI Ultra $199.99 and Claude Max $200 stay at 20x their base plans. Published GPT-6 Astra estimates in Codex run 5–45 local messages per 5 hours on Plus $20, 25–225 on Pro $100 and 100–900 on Pro $200 the 20x allowance, valid until 29 October . The verdict is to buy the plan that owns your agent.

watchResearch Gemini Ai Ultra Pricing

Google AI Ultra vs Claude Max vs ChatGPT Pro: Which $100 or $200 Plan for Developers? [UPDATED]

All three sell a 5x tier at $100 and a $200 tier, but ChatGPT Pro $200 drops to 10x on 30 Oct. We compare coding agents, limits and bundles by persona.

Sep 24 · 14 min