Vin ./ retweeted
interesting chart from @Gradient_HQ data's zero2repo
ai coding is no longer about writing code
it's about completing entire projects from scratch
and wen the same benchmark costs anywhere from ~$3 to $200 depending on the model and harness, efficiency becomes a pretty serious question
the smartest model isn't always the smartest way to spend your money
SWE-bench Verified has become saturated. Seven models scored 95–97% (Vals AI, Sep 2026). But how well can coding agents build complete software projects from scratch?
Zero2Repo is a Gradient Data research initiative building hard, verifiable tasks and training data for whole-project development.
Agents start with a full spec, an interface contract and an empty workspace. They build the entire repo without reference code, then submit it for hidden deterministic testing. Every test must pass for a task to count. No LLM judge is used.
The first release includes 17 tasks derived from real open-source projects across Python, TypeScript, Go, C, C++ and Zig.
Across 18 model–harness configurations, scores range from 1/17 to 15/17. The top 10 span 47 percentage points, compared with 5 points on SWE-bench Verified.
No configuration has completed all 17 tasks. This matters for self-iteration: small unresolved issues can persist and compound as agents continue developing a project. Reliable self-evolution remains an open research challenge.
We’re expanding the task collection and looking for contributors to suggest seed repositories, develop tasks and review requirements and hidden tests.
More cases are coming.
👉 zero2repo.ai
👉 github.com/Gradient-Data/Zer…
Vin ./ retweeted
lower prices and efficiency from open models keeping the utilization high on GPUs and builders.
commoditization in effect.
The "Rack Spread" is unfolding, token prices are falling while GPU prices climb
Change since mid-July:
- B300 GPU rentals up 106%
- B200 up 66%
- H100 up 17%
- Average token prices down 58%
Charts of the Week: a16z.news/p/charts-of-the-we…
Vin ./ retweeted
DeepSeek V4.1 Flash (1017) solved 8/17 Zero2Repo tasks with dsh, at $3.18 in reported API spend across the full suite.
OpenCode scored 7/17 and Claude Code, 5/17.
Costs are reported API spend across all 17 tasks.
Check out more findings: zero2repo.ai/
Vin ./ retweeted
Over the past year, we’ve worked on Echo and Logits to make AI post-training more efficient and accessible.
Gradient Data now extends that work with verifiable environments, real-world trajectories and evaluation systems for more capable and reliable agents.
gradientdata.ai
Vin ./ retweeted
Can web agents learn through imagination?
DynaWeb: Model-Based Reinforcement Learning of Web Agents has been accepted to EMNLP 2026 Main Conference.
Training web agents with online RL means costly, risky interaction with the live web. DynaWeb learns a Web World Model instead, letting agents dream web interactions and train on imagined rollouts.
- Learns naturalistic web transitions from agent actions
- Generates imagined trajectories in a synthetic web environment
- Interleaves imagined rollouts with real expert trajectories
- Consistent gains for open-source agents on WebArena and WebVoyager
Imagination as a scalable path toward online agentic RL.
This is part of our ongoing work at Gradient on distributed and efficient RL: training capable agents without the cost, latency, and risk of centralizing them.
Paper: arxiv.org/abs/2601.22149
Code: github.com/jadeleiyu/MBRL-Ag…
Vin ./ retweeted
Agents are impressive. Multi-agent systems are where things get interesting.
As AI systems evolve from single agents into networks of specialists, a new question follows:
How do agents decide who does what?
That’s a problem we’ve been exploring with Symphony.
Symphony brought multi-agent collaboration across heterogeneous environments. To take that further, we released Symphony-Coord, where coordination and specialization adapt through task context, performance, and feedback.
github.com/GradientHQ/sympho…
It's been rough lately, but it was all worth it in the end. AI Taiwan Expo was a blast, i had been able to connect with some wonderful people and learnt quite a lot.
This will be one of the core memory of 2026 for sure.
相信自己,相信您的理想。
Special thanks to Dexter, btw^^!
2
8
291
Vin ./ retweeted
The Bitter Lesson of Asynchronous RLHF:
Why Staleness Matters and How to Control It linkedin.com/pulse/bitter-le…
Vin ./ retweeted
A self-evolving agent + a 428B model + 3 Macs = ?
Your own AI lab.
We ran @MiniMax_AI M3 locally with @tryParallax, right on our desk.
Then @GA_agent_ai took over to create a 5-stock portfolio and write it to disk.
No cloud. No API bills. Nothing left the machine.
Wild to see a ~3K-line agent drive all this with a 400B+ model on local hardware.
Thanks to the GenericAgent and MiniMax teams for making local AI feel real.
Vin ./ retweeted
Fraction of the bill. Same results.
Fully local, open source, works with any client.
Just > pipx install uncommon-route
github.com/CommonstackAI/Unc…
Uncommonroute - automatic local router for api savings
Deepseek V4 available on @commonstack_ai within a day !!
The update is as fast as the throughput of the api!
DeepSeek V4 flash is on par with GPT 5.4 (high), the best part is that it’s much more affordable at scale:
GPT 5.4 pro vs DeepSeek V4 flash:
Input: $30/M vs $0.14/M (214x cost difference)
Output: $180/M vs $0.28/M (643x cost difference)
Both at a million context, DeepSeek V4 Flash is really a bargain for intelligence.
7
205
Finally, @tryParallax getting the attention it deserves.
You can run your own too at a fraction of the cost and it's private
Awesome to see @tryParallax’s distributed framework for heterogeneous machines being implemented and serving up inferences!
Build and customize your own clusters for AI like never before 🤖
./ LFG @Gradient_HQ
Vin ./ retweeted
Our cofounder @0xEricYang sat down with @yacinelearning to walk through Echo-2’s distributed RL architecture.
Dive in to learn about async RL with distributed infra, and how we are scaling this for businesses to win in the agentic era.
for those interested in distributed reinforcement learning I just finished a ~1h tutorial on the echo2 framework by @Gradient_HQ
we check:
- how to do async RL
- infra split between rollout workers and centralized learner
- interview with gradient cofounder eric yang himself!
Vin ./ retweeted
for those interested in distributed reinforcement learning I just finished a ~1h tutorial on the echo2 framework by @Gradient_HQ
we check:
- how to do async RL
- infra split between rollout workers and centralized learner
- interview with gradient cofounder eric yang himself!























