AIQB
TutorialsOrdinary

An 8B Agent That Thinks Without Writing Tokens Beats GPT-5.6 Luna on GAIA. Not on HLE

Source: DEV Community·
GPT

Summary

HybridAgent reasons partly in hidden states, then acts in text. Open weights, MIT. The wins are real, so are losses, and your traces can't read its thoughts.
CategoryAI Tutorials & Practice
TierOrdinary
Published
Indexed by AIQB
SourceDEV Community
AIQB record IDintel-1479bf866fdd61d2a6c09b0a
Read original source ↗More Tutorials

Related stories

Matched by category, entity tags and source

TutorialsOrdinary
Codex Swapped Its Default Model and Retired GPT-5.5 in 15 Days. Did Your Evals Notice?
Codex made GPT-6.1 Sol its default on Sep 29 and dropped GPT-5.5 for ChatGPT sign-ins on Oct 14. Pin your agent's model and log it with every result.
Source: DEV Community · ChatGPT / GPT
TutorialsOrdinary
ChatGPT UTM Bug Sent Referral Traffic to Direct During the GPT-6 Rollout
A temporary change in how ChatGPT handled outbound links distorted a key marketing metric during the...
Source: DEV Community · ChatGPT / GPT
TutorialsOrdinary
Gemini Gems become skills in November and custom GPTs retire on December 11. What to copy out first.
If you built a Gem in Gemini or a custom GPT in ChatGPT, it is about to be converted into something...
Source: DEV Community · ChatGPT / GPT / Gemini
TutorialsOrdinary
Why did my Claude Haiku calls start 400ing? Three AI API changes that break code quietly
Three AI API changes from the last two weeks that quietly break shipped code: Claude Haiku 5.5 migration 400s, Managed Agents web_fetch gating, and the GPT-6 plan split.
Source: DEV Community · GPT / Claude
TutorialsOrdinary
My benchmark scored GPT-5.4 mini 1.00 by silently skipping the 3 questions it failed
This is a submission for the Kaggle Benchmarking Challenge GPT-5.4 mini got a perfect 1.00 on my...
Source: DEV Community · GPT
TutorialsOrdinary
GPT-2 is the transformer plus five decisions, and inference is a different program
Originally published at cchinchilla.dev. Part 6 of From code to weights, a 12-part series on ML...
Source: DEV Community · GPT
TutorialsOrdinary
Spellcheck can't fix "chai GPT": building an LLM caption checker on $5 a month
Here's a line from YouTube's auto-captions on a lecture about attention in neural...
Source: DEV Community · GPT
TutorialsOrdinary
Same prompt, four models: what Opus, Sonnet, Astra and Sol each got wrong
I ran one prompt through four models last week. Opus 5.5 and Sonnet 5.5 in Claude Code, GPT-6 Astra...
Source: DEV Community · GPT / Claude
TutorialsOrdinary
Opus 5.5 pricing vs GPT-6 Sol: which is cheaper per task?
Opus 5.5 pricing: $4/$20 per million tokens, 60% off cache reads. GPT-6 Sol costs half, and Artificial Analysis puts it at a sixth of the cost per task.
Source: DEV Community · GPT
TutorialsOrdinary
OpenAI DevDay 2026: every announcement, with prices and availability
OpenAI DevDay 2026 recap: dots, GPT-6.1 Sol at $2/$10, Ultrafast, Pro 500, Decisions API, Codex cloud, Agents API and Sign in with ChatGPT.
Source: DEV Community · OpenAI / ChatGPT / GPT