Deepseek has released GPT-5.6-Sol except it costs $0.006 (cache in)/$0.3 (cache write)/$1.2 (out) and it runs at 200 fucking TPS and has a TTFT of a second.
Oh and also sometimes its half price.
Oh and also sometimes its half price.
dear god
September 10, 2026 at 6:40 AM
Deepseek has released GPT-5.6-Sol except it costs $0.006 (cache in)/$0.3 (cache write)/$1.2 (out) and it runs at 200 fucking TPS and has a TTFT of a second.
Oh and also sometimes its half price.
Oh and also sometimes its half price.
Woke up ready for Through The Fly Thursday in my new Amazon Essentials TW Briefs
#ttft #throughtheflythursday #hardgaycock #tightywhities #gaybriefs #cutgaycock
#ttft #throughtheflythursday #hardgaycock #tightywhities #gaybriefs #cutgaycock
January 9, 2025 at 3:52 PM
Woke up ready for Through The Fly Thursday in my new Amazon Essentials TW Briefs
#ttft #throughtheflythursday #hardgaycock #tightywhities #gaybriefs #cutgaycock
#ttft #throughtheflythursday #hardgaycock #tightywhities #gaybriefs #cutgaycock
Playing a little VR in my Stanfords on this Through The Fly Thursday!
#ttft #throughtheflythursday #gaycock #gayundiesfetish #gaymerpup #cutcock #gaybriefs #tightywhities
#ttft #throughtheflythursday #gaycock #gayundiesfetish #gaymerpup #cutcock #gaybriefs #tightywhities
December 27, 2024 at 12:47 AM
Playing a little VR in my Stanfords on this Through The Fly Thursday!
#ttft #throughtheflythursday #gaycock #gayundiesfetish #gaymerpup #cutcock #gaybriefs #tightywhities
#ttft #throughtheflythursday #gaycock #gayundiesfetish #gaymerpup #cutcock #gaybriefs #tightywhities
TTFT @aistudio.bsky.social
#QP Wild Spirit #aiart #digitalart
#generativeart #synthart #kazeai
#WolfWednesday #WolfAndHer
#QP Wild Spirit #aiart #digitalart
#generativeart #synthart #kazeai
#WolfWednesday #WolfAndHer
March 25, 2026 at 3:45 PM
February 19, 2026 at 11:20 PM
What Is TTFT? Understanding Time to First Token in LLMs
TTFT (Time to First Token) measures the time elapsed from a request to the first output token, not the overall speed of a large language model.
#hackernews #llm #news
TTFT (Time to First Token) measures the time elapsed from a request to the first output token, not the overall speed of a large language model.
#hackernews #llm #news
What Is TTFT? Understanding Time to First Token in LLMs
TTFT (Time to First Token) measures the time elapsed from a request to the first output token, not the overall speed of a large language model.
hackernoon.com
September 24, 2026 at 11:54 PM
What Is TTFT? Understanding Time to First Token in LLMs
TTFT (Time to First Token) measures the time elapsed from a request to the first output token, not the overall speed of a large language model.
#hackernews #llm #news
TTFT (Time to First Token) measures the time elapsed from a request to the first output token, not the overall speed of a large language model.
#hackernews #llm #news
#QP Blueprint of Advanced Tactical Starship
Inspired by @modean987.bsky.social
TTFT @nevyn79.bsky.social
Inspired by @modean987.bsky.social
TTFT @nevyn79.bsky.social
June 24, 2026 at 4:11 PM
#QP Blueprint of Advanced Tactical Starship
Inspired by @modean987.bsky.social
TTFT @nevyn79.bsky.social
Inspired by @modean987.bsky.social
TTFT @nevyn79.bsky.social
#QP Windblown Hair-Linen Shirt-Seaside Cliff
#aiart #3WordChallenge by @caylacultofai.bsky.social @joyaz.bsky.social
TTFT @aistudio.bsky.social
#aiart #3WordChallenge by @caylacultofai.bsky.social @joyaz.bsky.social
TTFT @aistudio.bsky.social
January 24, 2026 at 8:49 AM
#QP Windblown Hair-Linen Shirt-Seaside Cliff
#aiart #3WordChallenge by @caylacultofai.bsky.social @joyaz.bsky.social
TTFT @aistudio.bsky.social
#aiart #3WordChallenge by @caylacultofai.bsky.social @joyaz.bsky.social
TTFT @aistudio.bsky.social
THAT SAID... with kv caching and ttft under 500ms, "real feel" absolutely is around 45-50 tps. It does not feel "slow" by any means.
Increased Bonzai 2 27b's tokens per second by 30% by enabling KV caching.
Haven't been able to hit 50 tps yet... but there's still some juice left to squeeze here.
Having to use a fork of Prism ML's llama.cpp (AMD) isn't helping either, as there are no speculative decoders available.
Haven't been able to hit 50 tps yet... but there's still some juice left to squeeze here.
Having to use a fork of Prism ML's llama.cpp (AMD) isn't helping either, as there are no speculative decoders available.
September 20, 2026 at 7:25 PM
THAT SAID... with kv caching and ttft under 500ms, "real feel" absolutely is around 45-50 tps. It does not feel "slow" by any means.
Getting 124 tps and instantaneous ttft on a 64g m4. FLYING.
December 27, 2025 at 11:03 PM
Getting 124 tps and instantaneous ttft on a 64g m4. FLYING.
We studied time to first token (TTFT) and how it scales with increasing context length for GPT and Claude models. We found a significant difference, with GPT showing a noticeable quadratic component, while Claude models remain closer to linear.
September 8, 2026 at 9:12 PM
We studied time to first token (TTFT) and how it scales with increasing context length for GPT and Claude models. We found a significant difference, with GPT showing a noticeable quadratic component, while Claude models remain closer to linear.
Quick review:
The model ALWAYS thinks about using tools. Every reasoning block I get from it has it states that it's decided to use no tools. Overfit for agent stuff?
It's very very fast. This was from Parasail. TTFT was not great, however. A few seconds at least.
Fails the carwash test
The model ALWAYS thinks about using tools. Every reasoning block I get from it has it states that it's decided to use no tools. Overfit for agent stuff?
It's very very fast. This was from Parasail. TTFT was not great, however. A few seconds at least.
Fails the carwash test
April 7, 2026 at 3:48 AM
Quick review:
The model ALWAYS thinks about using tools. Every reasoning block I get from it has it states that it's decided to use no tools. Overfit for agent stuff?
It's very very fast. This was from Parasail. TTFT was not great, however. A few seconds at least.
Fails the carwash test
The model ALWAYS thinks about using tools. Every reasoning block I get from it has it states that it's decided to use no tools. Overfit for agent stuff?
It's very very fast. This was from Parasail. TTFT was not great, however. A few seconds at least.
Fails the carwash test
I think they had Copilot write the blog post too
July 31, 2024 at 9:53 PM
I think they had Copilot write the blog post too
yeah i mean that's fine i don't really care about latency anyway
July 4, 2026 at 7:17 PM
yeah i mean that's fine i don't really care about latency anyway
but like im getting 3 a ttft on a clean context vs 60s with ollama
May 23, 2026 at 4:22 PM
but like im getting 3 a ttft on a clean context vs 60s with ollama
🚀 Gemma4 Java runner:
• Single file, no deps
• E2B→31B + MoE
• GGUF + quant (F16→Q8)
• Vector API ⚡
• CLI + thinking modes
• GraalVM native + instant TTFT
Pure Java 😏
by @mukel.bsky.social
github.com/mukel/gemma4...
• Single file, no deps
• E2B→31B + MoE
• GGUF + quant (F16→Q8)
• Vector API ⚡
• CLI + thinking modes
• GraalVM native + instant TTFT
Pure Java 😏
by @mukel.bsky.social
github.com/mukel/gemma4...
github.com
April 9, 2026 at 7:12 AM
🚀 Gemma4 Java runner:
• Single file, no deps
• E2B→31B + MoE
• GGUF + quant (F16→Q8)
• Vector API ⚡
• CLI + thinking modes
• GraalVM native + instant TTFT
Pure Java 😏
by @mukel.bsky.social
github.com/mukel/gemma4...
• Single file, no deps
• E2B→31B + MoE
• GGUF + quant (F16→Q8)
• Vector API ⚡
• CLI + thinking modes
• GraalVM native + instant TTFT
Pure Java 😏
by @mukel.bsky.social
github.com/mukel/gemma4...
cheapestinference.com just increased prices. Maybe it still sounds like a good offer, but beware that performance is fluctuating wildly. While often OK, TTFT goes up to over a minute for extended periods of time.
Cheapest AI Inference — Unlimited LLM API, flat $22/mo
Cheapest AI inference pricing: unlimited Kimi K3, Qwen3.8 Max & more from $22/mo flat — no per-token billing. Works with Claude Code.
cheapestinference.com
September 23, 2026 at 8:06 PM
cheapestinference.com just increased prices. Maybe it still sounds like a good offer, but beware that performance is fluctuating wildly. While often OK, TTFT goes up to over a minute for extended periods of time.
Run a local LLM like a service (HTTP), stream NDJSON, and measure TTFT + throughput. Includes Node/Python/Go/C++ streaming clients.
methodicalfunction.com/log/2026/01/...
#Ollama #LocalLLM #DevTools
methodicalfunction.com/log/2026/01/...
#Ollama #LocalLLM #DevTools
Run a Local LLM Like a Service: Streaming + TTFT Metrics
Run an Ollama model locally on macOS, Linux, or Windows, stream output over HTTP, and measure TTFT and throughput with Node, Python, Go, and C++.
methodicalfunction.com
January 22, 2026 at 7:18 PM
Run a local LLM like a service (HTTP), stream NDJSON, and measure TTFT + throughput. Includes Node/Python/Go/C++ streaming clients.
methodicalfunction.com/log/2026/01/...
#Ollama #LocalLLM #DevTools
methodicalfunction.com/log/2026/01/...
#Ollama #LocalLLM #DevTools
TTFT (Time to First Token) measures the time elapsed from a request to the first output token, not the overall speed of a large language model. #aiinfrastructure
What Is TTFT? Understanding Time to First Token in LLMs
hackernoon.com
September 24, 2026 at 5:35 AM
TTFT (Time to First Token) measures the time elapsed from a request to the first output token, not the overall speed of a large language model. #aiinfrastructure
#原創
「聽聞這梁國公府的嫡庶姐妹感情不很好啊？」第五篇！
接下來幾篇都是垃圾未婚夫與綠茶白月光，嗚呼！
然後照例奉上柳柳顏藝小圖，之後有機會也來幾個婉婉的❤️
第一篇：
www.facebook.com/share/p/2FoD...
第二篇：
www.facebook.com/share/p/Tgby...
第三篇：
www.facebook.com/share/p/tTft...
第四篇：
www.facebook.com/share/p/1W9A...
「聽聞這梁國公府的嫡庶姐妹感情不很好啊？」第五篇！
接下來幾篇都是垃圾未婚夫與綠茶白月光，嗚呼！
然後照例奉上柳柳顏藝小圖，之後有機會也來幾個婉婉的❤️
第一篇：
www.facebook.com/share/p/2FoD...
第二篇：
www.facebook.com/share/p/Tgby...
第三篇：
www.facebook.com/share/p/tTft...
第四篇：
www.facebook.com/share/p/1W9A...
November 3, 2024 at 10:42 PM
#原創
「聽聞這梁國公府的嫡庶姐妹感情不很好啊？」第五篇！
接下來幾篇都是垃圾未婚夫與綠茶白月光，嗚呼！
然後照例奉上柳柳顏藝小圖，之後有機會也來幾個婉婉的❤️
第一篇：
www.facebook.com/share/p/2FoD...
第二篇：
www.facebook.com/share/p/Tgby...
第三篇：
www.facebook.com/share/p/tTft...
第四篇：
www.facebook.com/share/p/1W9A...
「聽聞這梁國公府的嫡庶姐妹感情不很好啊？」第五篇！
接下來幾篇都是垃圾未婚夫與綠茶白月光，嗚呼！
然後照例奉上柳柳顏藝小圖，之後有機會也來幾個婉婉的❤️
第一篇：
www.facebook.com/share/p/2FoD...
第二篇：
www.facebook.com/share/p/Tgby...
第三篇：
www.facebook.com/share/p/tTft...
第四篇：
www.facebook.com/share/p/1W9A...
as cool as local-sized models are I don't understand why some people are so fixated on them as the only alternative to big labs
there are options between "10min TTFT on my $6000 macbook" and "the US govt must hand us a 12-month lead over open-source for Safety"
there are options between "10min TTFT on my $6000 macbook" and "the US govt must hand us a 12-month lead over open-source for Safety"
May 21, 2026 at 9:22 PM
as cool as local-sized models are I don't understand why some people are so fixated on them as the only alternative to big labs
there are options between "10min TTFT on my $6000 macbook" and "the US govt must hand us a 12-month lead over open-source for Safety"
there are options between "10min TTFT on my $6000 macbook" and "the US govt must hand us a 12-month lead over open-source for Safety"