Releases·Tool Release
Astorias AI launches cheap inference hosting Qwen3.8 27B, plans Agent Sandbox
Astorias AI launched an inference platform aimed at affordable pricing and reliable agent tooling, debuting with Qwen 3.8 27B. The founder shared pricing and solicited interest in a planned serverless "Agent Sandbox" feature. GLM 5.3 Flash and other flash models are planned next.
4 engagement1 sourceTue, Sep 22, 2026, 11:51 AM
AI Overview
- Astorias AI prices Qwen3.8 27B at $0.1/M input, $1.6/M output, $0.05/M cached tokens, calling it among the cheapest
- Planned "Agent Sandbox" would offer serverless containers, e.g. 8GB 4vCPU at $0.015, undercutting Novita and Modal
- Astorias plans to add GLM 5.3 Flash and other flash models at affordable prices
Related
Releases·Tool Release
ObitMC launches Rust-based inference engine running Qwen 3.8 27B on Spot compute, ~65% cheaper than OpenRouter
Founder Julian launched ObitMC, an inference startup whose custom Rust scheduler and inference engine runs Qwen 3.8 on interruptible/unreliable compute (AWS/GCP Spot, Vast Interruptible), claiming ~65% savings versus OpenRouter. He is offering 100 million free tokens (~$12 credit) to test the service, with ongoing pricing of $0.10/M input, $1/M output, and $0.01/M cached. The offer targets real production load testing rather than serving as a promotion.
Founder Julian launched ObitMC, an inference startup whose custom Rust scheduler and inference engine runs Qwen 3.8 on interruptible/unreliable compute (AWS/GCP Spot, Vast Interruptible), claiming ~65% savings versus OpenRouter. He is offering 100 million free tokens (~$12 credit) to test the service, with ongoing pricing of $0.10/M input, $1/M output, and $0.01/M cached. The offer targets real production load testing rather than serving as a promotion.
1 engagement1 sourceredditThu, Oct 1, 2026, 07:19 AM
Social·Community
cheapest way to run agent loops while learning in 2026, free tiers or cheap pay per token
A Reddit user asks about the cheapest usable setups for learning to build real agent loops (tools, memory, long contexts) in 2026, arguing free tiers die quickly under repeated runs. They report the cheap end is DeepSeek V4 Flash (0731) at roughly six cents and GLM 5.3 Flash, with pay-per-token beating subscriptions.
A Reddit user asks about the cheapest usable setups for learning to build real agent loops (tools, memory, long contexts) in 2026, arguing free tiers die quickly under repeated runs. They report the cheap end is DeepSeek V4 Flash (0731) at roughly six cents and GLM 5.3 Flash, with pay-per-token beating subscriptions.
6 engagement1 sourceredditMon, Sep 21, 2026, 11:55 AM
Social·Community
GLM 5.3 Flash beat DeepSeek V4 Flash Vision Exp in a browser screenshot QA test
A developer running a browser QA harness that converts screenshots into actions compared DeepSeek V4 Flash Vision Exp and GLM 5.3 Flash, citing cost concerns with prior vision models at high screenshot volumes. GLM performed better on their test set. Both models ship under the MIT license with hosted inference options.
A developer running a browser QA harness that converts screenshots into actions compared DeepSeek V4 Flash Vision Exp and GLM 5.3 Flash, citing cost concerns with prior vision models at high screenshot volumes. GLM performed better on their test set. Both models ship under the MIT license with hosted inference options.
8 engagement1 sourceredditFri, Sep 4, 2026, 09:51 AM
Releases·Tool Release
openjev-sglang: Qwen3.6-35B-A3B prefill-only server on SGLang, deployed on Modal B200s with TypeSafe/Jev API
A GitHub project, openjev-sglang, implements the TypeSafe/Jev HTTP API as a prefill-only endpoint running Qwen3.6-35B-A3B on SGLang, deployable on Modal with one B200 per container. It matters to practitioners because it offers an open-model substitute for the proprietary Jev API with one-command Modal deployment and scale-to-zero cost control.
A GitHub project, openjev-sglang, implements the TypeSafe/Jev HTTP API as a prefill-only endpoint running Qwen3.6-35B-A3B on SGLang, deployable on Modal with one B200 per container. It matters to practitioners because it offers an open-model substitute for the proprietary Jev API with one-command Modal deployment and scale-to-zero cost control.
82 engagement1 sourcegithubThu, Sep 17, 2026, 06:11 PM
Developments·Benchmark
Qwen 3.8 27B matches Gemini 2.5 Pro on Aider benchmark
A user benchmarked Qwen 3.8 27B FP8 with FP8 KV cache and 256K context on the Aider coding benchmark, scoring 72.9. This matches Gemini 2.5 Pro's score from April 2025 and beats Claude Opus 4 and DeepSeek R1, highlighting the model's strong coding performance on a MacBook.
A user benchmarked Qwen 3.8 27B FP8 with FP8 KV cache and 256K context on the Aider coding benchmark, scoring 72.9. This matches Gemini 2.5 Pro's score from April 2025 and beats Claude Opus 4 and DeepSeek R1, highlighting the model's strong coding performance on a MacBook.
74 engagement1 sourceredditMon, Aug 24, 2026, 07:56 AM