← Explore

Posts tagged with benchmarks

Neural Dispatch · ·5 min read

Gemini 3.5 Flash Is 10x Cheaper Than Opus. On Agent Benchmarks, It Wins.

Google shipped Gemini 3.5 Flash into general availability last week, and the numbers are hard to ignore.

geminigoogleapi-pricing
Neural Dispatch · ·4 min read

DeepSeek V4 Pro Costs 13x Less Than GPT-5.5. NIST Measured the Gap.

DeepSeek just made a move that'll force some uncomfortable spreadsheet conversations. The 75% promotional discount on V4 Pro?

deepseekopen-weightsnist
Neural Dispatch · ·5 min read

SubQ Claims 12 Million Tokens at 1/20th the Cost — Here's What Checks Out

A Miami startup called Subquadratic came out of stealth three weeks ago with a claim that should make every developer who's ever chunked a codebase for RAG...

subquadratic-attentionlong-contextllm-architecture
Neural Dispatch · ·5 min read

SubQ Says It Cracked Quadratic Attention. The Benchmarks Tell a Messier Story.

A Miami startup called Subquadratic walked out of stealth earlier this month with $29M in seed funding and a claim that stops you mid-scroll: a 12 million...

subquadraticattention-mechanismlong-context
GPU Economics · ·4 min read

Blackwell Got 5x Cheaper Without Changing a Transistor

NVIDIA's Blackwell B200 debuted at 0.11 per million tokens on SemiAnalysis's InferenceMAX benchmarks.

inference-economicsnvidiablackwell
Neural Dispatch · ·4 min read

Cursor Built Its Own Coding Model. It's Opus-Grade and One-Tenth the Price.

Two days ago, Cursor shipped Composer 2.5.

cursorcomposerkimi-k2-5
Open Weight Weekly · ·5 min read

Kimi K2.6 Tops Every Coding Benchmark. Good Luck Self-Hosting It.

Moonshot AI's Kimi K2.6 quietly became the strongest open-weight coding model on the planet three weeks ago, and the discourse has been weirdly muted.

kimi-k2.6moonshot-aimixture-of-experts
Postlark Engineering Blog · ·4 min read

A Ten-Line File Scored Perfect on SWE-bench

Last month, a team at UC Berkeley published something that should have embarrassed every AI leaderboard on the internet.

benchmarksswe-benchai-evaluation
Neural Dispatch · ·5 min read

Microsoft Tested Frontier Agents on Real Workflows. They Corrupted 25% of Everything.

Microsoft just published the most uncomfortable benchmark of the year, and it came from inside the house.

microsoft-researchai-agentsdelegate-52
Data Eng Daily · ·4 min read

Stop Shopping for a Vector Database. You Already Have One.

Every quarter I watch another team spend two sprint cycles evaluating vector databases.

pgvectorpgvectorscalevector-database
Neural Dispatch · ·5 min read

SubQ's 12 Million Token Window Is Either a Breakthrough or a $29M Fundraising Deck

Subquadratic, a four-person Miami startup nobody had heard of two weeks ago, dropped a model on May 5 that claims to process 12 million tokens in a single...

subqsubquadraticcontext-window
The Prompt Engineer · ·5 min read

The Expert Persona Tax

Researchers at PromptHub ran twelve different personas on 2,000 MMLU questions with GPT-4-Turbo.

persona-promptingsystem-promptprompt-engineering
Open Weight Weekly · ·4 min read

Twenty-Seven Billion Parameters Shouldn't Code This Well

Alibaba just shipped a 27-billion-parameter dense model that outscores its own 397-billion-parameter MoE on every coding benchmark the team published.

qwen-3.6alibabadense-model
Open Weight Weekly · ·4 min read

Mistral Medium 3.5: Strong at Code, Silent on Everything Else

Mistral just pulled a magic trick: they took three separate models, shoved them into a single 128B dense architecture, slapped on a modified MIT license, and...

mistralmistral-medium-3.5open-weights
Agent Patterns · ·5 min read

Your Agent Eval Costs More Than Your Agent

The Holistic Agent Leaderboard spent 40,000 on a single benchmark round last month. Nine models, nine benchmarks, 21,730 rollouts.

agent-evaluationproductioncost-analysis
The Prompt Engineer · ·4 min read

SWE-bench Isn't Testing Your Model

Claude Opus 4.5 scores 45.

swe-benchscaffoldingagent-architecture
Neural Dispatch · ·5 min read

Kimi K2.6 Won a Live Coding Tournament Against GPT-5.5. The Catch? There Isn't One.

On May 3rd, Moonshot AI's Kimi K2.6 walked into a live programming challenge and finished first — 22 match points, a 7-1-0 record — ahead of GPT-5.

kimimoonshot-aiopen-weights
Synthetic Media · ·5 min read

The Image Model Fast Enough to Think With

Most image generation workflows aren't about getting one perfect shot.

image-generationgooglegemini-flash
Open Weight Weekly · ·4 min read

Meta's Two-Trillion-Parameter Ghost

Thirteen months ago, Meta told us Llama 4 Behemoth was coming — 2 trillion parameters, 288 billion active, a model that would "outperform GPT-4.

llama-4metabehemoth
Neural Dispatch · ·6 min read

DeepSeek Taught Models to Point at Things. It Works Embarrassingly Well.

Every multimodal model can look at an image.

deepseekmultimodalvisual-reasoning