Topic · 11 stories

#reasoning-models

Reporting, analysis and practical guides connected to reasoning-models.

Latest in #reasoning-models

Newest first

GRPO: The Critic-Free RL Algorithm Behind DeepSeek-R1

GRPO (Group Relative Policy Optimization) is a critic-free reinforcement learning algorithm introduced in the DeepSeekMath paper (arXiv 2402.03300). Instead of training a separate value model like PPO, it samples a group of responses per prompt and computes each response's advantage relative to the group's mean and standard deviation. It powered DeepSeek-R1's emergent reasoning and is the central baseline for reinforcement learning with verifiable rewards in 2026, spawning variants like Dr. GRPO, DAPO, and GSPO.

Jul 22, 2026 · 6 min read

AI News

Grok 4.5: xAI's Opus-Class Coder at a Third of the Price

Grok 4.5, released July 8, 2026, is xAI's coding-focused model. It ranks 4th on the Artificial Analysis Intelligence Index (score 54), wins SWE Marathon (29%), and prices at $2/$6 per million tokens with 4.2x better token efficiency than Opus 4.8. Not yet available in the EU.

Jul 12, 2026 · 5 min read

AI News

GPT-Realtime-2.1: OpenAI Adds Reasoning to Its Voice API

On July 6, 2026, OpenAI released GPT-Realtime-2.1 and GPT-Realtime-2.1-mini for the Realtime API. The headline change is reasoning in the low-cost mini tier, plus a 25% cut in p95 latency from better caching. The mini holds the prior gpt-realtime-mini price (0 audio in, 0 audio out per 1M) while the full model runs 2/4. Reasoning effort is configurable from minimal to xhigh.

Jul 8, 2026 · 5 min read

AI News

GPT-5.6 Sol: OpenAI's Best Model, Held Back by Washington

On June 26, 2026, OpenAI previewed the GPT-5.6 series — Sol (flagship), Terra (balanced, 2x cheaper than GPT-5.5), and Luna (fastest, cheapest) — but restricted access to trusted partners at the US government's request due to the models' strong cybersecurity capabilities. OpenAI paired the release with its most robust layered safeguard stack and said it does not want government pre-release review to become the default.

Jul 2, 2026 · 6 min read

AI News

Claude Sonnet 5: Anthropic's Most Agentic Mid-Tier Model Yet

Claude Sonnet 5, released June 30, 2026, is Anthropic's most agentic mid-tier model. It beats Sonnet 4.6 on every published benchmark (63.2% SWE-bench Pro, 80.4% Terminal-Bench 2.1, 81.2% OSWorld) and edges Opus 4.8 on GDPval-AA v2 knowledge work. Intro pricing is /0 per million tokens through Aug 31, 2026, then /5. A new tokenizer can raise token counts up to 1.35x, and xhigh effort can cost more than Opus 4.8.

Jul 1, 2026 · 5 min read

AI News

MAI-Thinking-1: Microsoft's First In-House Reasoning Model

Microsoft unveiled MAI-Thinking-1 at Build 2026, its first reasoning model trained in-house without distillation. The 35B-active, ~1T-total MoE has a 256k context window, scores 97.0% on AIME 2025 and matches Claude Opus 4.6 on SWE-Bench Pro. It's in private preview on Microsoft Foundry.

Jun 23, 2026 · 5 min read

Deep Dives

Test-Time Compute: Why Reasoning Models Think Before Answering

Test-time compute spends extra computation during inference, not training, to improve answers. It powers reasoning models like OpenAI o1 and DeepSeek-R1. Two strategies exist: sequential scaling (longer chains of thought, e.g. the s1 paper's budget forcing) and parallel scaling (Best-of-N, majority voting). More thinking is not always better, overthinking degrades accuracy, and hidden reasoning tokens are billable. Match compute to task difficulty.

Jun 17, 2026 · 8 min read

Deep Dives

Diffusion LLMs: How Text Diffusion Is Challenging Autoregression

Diffusion language models (dLLMs) abandon left-to-right autoregressive generation, instead refining masked noise into text over a few parallel denoising steps. Inception Labs' Mercury Coder runs at 1,100+ tokens per second on H100s versus 50-200 for autoregressive models, and LLaDA 8B's bidirectional design breaks the reversal curse. They still trail the best models on hard reasoning benchmarks, but the one-token-at-a-time assumption is no longer a law of nature.

Jun 12, 2026 · 8 min read

Deep Dives

ZAYA1-8B: Zyphra's 760M-Active MoE Trained on AMD

Zyphra's ZAYA1-8B MoE model, trained on AMD, achieves high performance with efficient parameter activation.

May 24, 2026 · 6 min read

Open Source

Trinity-Large-Thinking: 400B U.S.-Made Open Reasoning Model

Trinity-Large-Thinking is Arcee AI's 400B open-weights reasoning model, offering powerful, cost-effective agent tuning.

Apr 30, 2026 · 7 min read

AI News

DeepSeek V4 Pro: 1.6T Open-Weights Model Hits #2 on the Index

DeepSeek V4 Pro is a top 1.6T open-weights model for agents, but has a high hallucination rate.

Apr 29, 2026 · 5 min read