Archive - Behind the Stack

Skip to content

Behind the Stack

Home
About
Contact
Archive

Search

Every post, grouped by year and month. Click a year or month to expand it.

2026 (10)

July (1)

GPT-5.6: a three-tier model family, a multi-agent mode, and a two-week government gateJul 14

June (2)

Claude Fable 5 Shipped, Then Got Pulled by an Export Control OrderJun 25
C# finally gets union types, and the compiler design tells you why it took this longJun 4

May (1)

GPT-5.5: A Ground-Up Rebuild Wearing a Point Release NumberMay 13

April (1)

Claude Opus 4.7: what actually changed under the hoodApr 21

March (2)

GPT-5.4: computer use stops being a bolt-onMar 31
Gemini 3.1 Pro and the price of abstract reasoningMar 9

February (1)

Claude Opus 4.6: What Actually Changed Under the HoodFeb 16

January (2)

Microsoft Foundry’s Agent Stack Grows Up: Memory, A2A, and the SDK v2 RewriteJan 25
NVIDIA Nemotron 3: A Hybrid Mamba-Transformer-MoE Stack Built for Agent Swarms, Not ChatbotsJan 3

2025 (16)

December (1)

Gemini 3: A Reasoning-First Model and Google’s Bet on Agentic IDEsDec 13

November (1)

GPT-5.1’s adaptive reasoning: pushing the routing decision inside the modelNov 21

October (2)

OpenAI’s Atlas browser and the OWL split: what the architecture actually buys youOct 31
Sora 2: What Actually Changed Under the HoodOct 9

September (1)

GPT-5-Codex and the shift toward dynamic-effort coding agentsSep 17

August (2)

GPT-5’s router problem is an architecture problemAug 27
Kimi K2: A Trillion-Parameter Open Model Built Around Agentic BehaviorAug 5

July (1)

Grok 4: What xAI’s Compute-Scaling Bet Actually Bought ThemJul 14

June (1)

OpenAI’s o3-pro: buying reliability with compute, not a new modelJun 22

May (2)

Claude Opus 4 and Sonnet 4: What Actually Changed Under the HoodMay 31
OpenAI’s o3 and o4-mini: Tool Calls Move Inside the Chain of ThoughtMay 10

April (1)

Meta ships Llama 4, then gets caught gaming the leaderboardApr 18

March (2)

Manus: What a Wrapped Agent Looks Like When the Wrapping Is Done WellMar 27
One model, two speedsMar 5

February (1)

o3-mini and the reasoning-effort knobFeb 12

January (1)

DeepSeek-R1: A Technical Look at the Reasoning Model That Skipped the SFT StepJan 21

2024 (17)

December (2)

o3 and the Test-Time Compute BetDec 30
A rebrand that actually changes the architectureDec 9

November (1)

.NET 9 shipped: what actually changed under the hoodNov 17

October (2)

Claude gets hands: inside Anthropic’s computer use releaseOct 26
OpenAI DevDay 2024: The Realtime API Changes How You Build Voice, and the Other Three Announcements Change How Much You PayOct 5

September (1)

OpenAI’s o1: a different scaling axisSep 13

August (2)

Grok-2 arrives, and takes xAI’s weights strategy in the opposite directionAug 22
Meta ships a 405B open-weight model that actually competesAug 1

July (1)

Claude 3.5 Sonnet: A Mid-Tier Model That Beat the FlagshipJul 10

June (1)

Apple Intelligence and Private Cloud Compute: A Look at the ArchitectureJun 18

May (2)

GPT-4o: What Changes When Voice, Vision, and Text Share One ModelMay 28
Llama 3: What Actually Changed Under the HoodMay 6

April (1)

Grok-1’s Open Weights Release: What xAI Actually ShippedApr 14

March (2)

Claude 3’s Three-Tier Launch: What Actually Changed Under the HoodMar 23
What actually shippedMar 2

February (1)

The GPT Store: What Actually Ships When You ‘Build a GPT’Feb 9

January (1)

Mixtral 8x7B: What a Sparse Mixture-of-Experts Model Actually Buys YouJan 18

2023 (17)

December (2)

Gemini 1.0: What Actually Shipped Versus What Was DemoedDec 28
The five days that exposed OpenAI’s structural fault lineDec 6

November (1)

OpenAI DevDay: GPT-4 Turbo and the Assistants API Change the Integration StoryNov 15

October (2)

Mistral 7B: A 7-Billion-Parameter Model That Beat 13B Llamas on Almost EverythingOct 24
ChatGPT Gets Eyes and Ears: What Actually Changed Under the HoodOct 2

September (1)

ChatGPT Enterprise: what actually changed under the hoodSep 10

August (1)

SDXL 1.0: What Actually Changed Under the HoodAug 19

July (2)

Llama 2: What Actually Changed Under the HoodJul 29
GPT-4’s API Graduates: General Availability, Function Calling, and What It Means for Production SystemsJul 7

June (1)

Falcon 40B takes the top of the Open LLM LeaderboardJun 15

May (2)

Azure AI Studio and the Copilot Stack: What Build 2023 Actually ShippedMay 24
Segment Anything: A Foundation Model for Image SegmentationMay 3

April (1)

AutoGPT and the autonomous agent loop: what actually changedApr 11

March (1)

GPT-4’s Launch: What Changed Under the Hood, What Didn’t Ship, and What It Means for BuildersMar 21

February (2)

Bing Chat’s Sydney Persona: A Postmortem on Grounding, System Prompts, and Prompt InjectionFeb 27
Azure OpenAI Service goes GA: what actually changedFeb 5

January (1)

Point-E: OpenAI’s Fast, Rough Answer to Text-to-3DJan 14

2022 (13)

December (2)

Stable Diffusion 2.0: what the CLIP swap actually broke, and whyDec 24
ChatGPT’s Launch: What’s Actually Different Under the HoodDec 2

November (1)

.NET 7 shipped. Here’s what actually changed for youNov 10

October (1)

Azure Cosmos DB for PostgreSQL: What the Citus GA Actually ChangesOct 20

September (2)

.NET 7 Hits Release Candidate: What Native AOT Actually ChangesSep 28
Stable Diffusion’s Public Release: What Changed When the Weights Went OutSep 6

August (1)

DALL-E 2 Goes to a Million Users: What Changed Wasn’t the ModelAug 15

July (2)

Midjourney’s Open Beta and the Case for Building on DiscordJul 24
What LaMDA Actually Is: A Technical Look at the Model Behind the Sentience ClaimJul 2

June (1)

GPT-3 gets an Azure resource groupJun 11

May (1)

One network, 604 tasksMay 20

April (2)

Inside PaLM: What Google’s 540B Model Actually ChangedApr 28
Inside DALL-E 2: How OpenAI’s unCLIP Architecture Actually WorksApr 7

© 2026 Behind the Stack

Azure, AI, and cloud engineering — from the trenches.

Designed with WordPress.