Model Pricing

Medical LVLMs are prone to factual inconsistencies and poor visual grounding. Existing alignment methods have three key limitations in the medical domain: sequence-level rewards treat clinically critical tokens equally, reliance on static SFT references causes off-policy shift, and alignment lacks visual grounding constraints. The proposed method uses a bidirectional token-wise KL regularizer and a visual-contrastive grounding objective, forming a fine-grained on-policy alignment framework that constructs preference pairs by minimally editing model outputs. Experiments validate its effectiveness.

OpenAI is considering cutting API token prices to win customers from Anthropic, according to the Wall Street Journal, signaling a potential price war in the AI industry.

Cohere has released its first developer-facing coding model, North Mini Code, a 30B total parameter mixture-of-experts model with only 3B active parameters per token. It runs on a single H100 GPU, supports 256K context length, and is optimized for code generation, agentic software engineering, and terminal tasks. The weights are open under Apache 2.0.

Google has released DiffusionGemma, a new open-weight model under Apache 2 license, available for free via NVIDIA's NIM cloud API. It delivers impressive generation speeds exceeding 500 tokens per second.

Google released DiffusionGemma, a 26-billion-parameter model that generates text via diffusion, achieving 1,000 tokens per second on an H100 GPU—four times faster than autoregressive models, but with lower quality. It's currently experimental.

DiffusionGemma is Google DeepMind's experimental open text generation model that uses text diffusion instead of standard autoregressive decoding, achieving up to 4x faster generation on dedicated GPUs. The 26B MoE model (3.8B active parameters) is built on the Gemma 4 backbone, supports multimodal inputs (text, image, video), has a 256K context window, covers 140+ languages, and is released under Apache 2.0.

Google DeepMind released DiffusionGemma, an experimental open model for fast text generation using parallel token generation. NVIDIA optimized it to run faster on GeForce RTX, RTX PRO, and DGX Spark systems, achieving up to 1000 tokens/sec locally.

Anthropic releases Claude Fable 5 and Mythos 5, with Fable 5 offering the same performance as Mythos 5 but with stricter safety guardrails. The models feature a 1 million token context window, 128k output tokens, and pricing double that of Opus 4.8. Simon Willison spent 5.5 hours testing Fable 5 and found it to be a 'beast'—knowledgeable and capable, but slow and expensive. Fable 5 successfully upgraded micropython-wasm to full Python, implemented pause-resume for tool calls in Datasette Agent and the LLM library, and consumed $110.42 in tokens in a single day.

Anthropic announced Claude Fable 5, claiming it is the most powerful AI model it has widely released, with exceptional performance in software engineering, knowledge work, and vision. It marks the first broad release from the Mythos class, previously deemed too dangerous due to cybersecurity capabilities. New safeguards block responses in high-risk areas, falling back to Claude Opus 4.8 when necessary. Anthropic also launched Claude Mythos 5, available only in a limited trusted-access program. Pricing is $10 per million input tokens and $50 per million output tokens.

Researchers identify a 'concept bottleneck' in CoCoNuT latent reasoning where intermediate hidden states are overwritten, causing performance loss. They propose AGCLR with a persistent gated memory (write, read, forget gates) that consistently improves performance on GSM8K, HotpotQA, and ProsQA using GPT-2, with the gap widening as curriculum depth increases.

AI-noleak is a local reverse proxy that intercepts accidentally exposed secrets (API keys, tokens) from AI coding agents and replaces them with deterministic placeholders before they reach the upstream AI model. It operates via three layers (PTY wrapper, HTTP proxy, file watcher) without requiring TLS MITM or root CA certificates, ensuring local security isolation.

This paper introduces On-Policy Diffusion Language Model (OPDLM), which transforms autoregressive models into diffusion language models via on-policy distillation, addressing distribution shifts. OPDLM achieves strong performance with 15x to 7,000x fewer training tokens across various tasks, positioning DLM transformation as a form of ARLM post-training.

The Piggyback Hypothesis proposes that chat-template tokens can piggyback finetuned behavior onto out-of-domain queries. Validated via prefix perturbations, leading to Token-Regularized Finetuning (TReFT) that mitigates emergent misalignment while preserving in-domain learning.

Diffusion Large Language Models (dLLMs) refine tokens iteratively but commit them irreversibly, causing a "stability lag" where early decisions remain fragile. Post-Training Quantization (PTQ) error easily flips these borderline decisions at the write frontier, locking them in. FAIR-Calib, a two-stage PTQ framework, probes a full-precision teacher for a position prior and performs off-policy layer-wise calibration with a reweighted hidden-state MSE, protecting fragile frontier states without expensive end-to-end rollouts. Theoretically justified as a surrogate for output KL divergence, FAIR-Calib outperforms baselines on LLaDA and Dream (W4A4), reducing frontier flips and post-commit mismatches.

This article details the configuration and memory calibration required to run the Qwen 3.6 35B MoE model at a 450,000 token context window on a single 32GB VRAM GPU (NVIDIA RTX 5090) using llama.cpp with TurboQuant and YaRN scaling. It covers model selection, quantization trade-offs, KV cache quantization, RoPE scaling, multimodal setup, replication guide, VRAM lifecycle management, and performance evaluation.

On June 3, 2026, Google introduced Gemma 4 12B Unified, an open-source multimodal model that understands text, images, audio, and video within a single architecture. It combines a 256K context window with a laptop-friendly design for agentic workflows and local deployment. This article covers its architecture, features, benchmarks, and practical guidance for developers.

NVIDIA Nemotron 3 Ultra, an open large language model with 550B total parameters and 55B active parameters, is now available on Amazon SageMaker JumpStart. It offers 5x faster inference and up to 30% lower cost for agentic AI workloads, with a hybrid Transformer-Mamba MoE architecture and million-token context window.

A comprehensive guide to AI API pricing, covering six key decisions: what to meter, which pricing primitive (tokens, credits, outcomes), cost calculation, tier structure, hard vs. soft caps, and credit wallet design. Includes practical examples and a diagnostic prompt for your own pricing.

Qwen 3.7 Plus is Alibaba's proprietary reasoning model released in June 2026, scoring 53 on the Artificial Analysis Intelligence Index, far above average. However, it is expensive, slow, and very verbose. The model supports text, image, and video input with a 1M-token context window.

Walmart is limiting employee use of its internal AI assistant Code Puppy by assigning fixed tokens due to high costs from the shift to pay-per-use LLM billing. The retailer aims to control expenses and encourage thoughtful AI usage.

GitHub Copilot has adopted a usage-based pricing system using credits. Costs vary by model and tokens, with advanced models being more expensive. Users report high credit consumption even for simple tasks, and caution is needed with Auto mode.

MiniMax officially released MiniMax M3 on June 1, 2026, featuring MiniMax Sparse Attention (MSA) for a 1M-token context window, native image/video input, and desktop computer operation. The API is live now.

This post explores how combining Amazon FSx for Lustre, NVIDIA GPUDirect Storage, and sharded parallel loading reduces cold-start time-to-first-token for large language models from minutes to seconds, and how TurboQuant KV cache significantly increases context window size.

Chinese AI company MiniMax has released its new model M3. It's billed as the first open-weight model to combine top-tier coding performance, a one-million-token context window, and native multimodality.

NVIDIA announced Nemotron 3 Ultra, a 550B-parameter open weights model with 55B active parameters, achieving the highest intelligence among US open weights models with a score of 48 on the AI Index, and serving over 300 tokens per second on DeepInfra.

Chinese AI startup MiniMax released its flagship model M3, designed for coding agents and automated workflows. It processes up to 1M tokens, reduces computational costs by 20x, and outperforms OpenAI GPT-5.5 and Google Gemini on SWE-Bench Pro. The company also prepares for a Shanghai IPO and partners with Ant Group's Alipay for AI payment infrastructure.

At the 2026 China AIGC Industry Summit, Wang Xiaoye, Technical Director of AWS Product Technology, pointed out that 87% of enterprises claim to have deployed AI at scale, but only 10% have gained actual value. He emphasized the huge gap between personal and enterprise-level agent deployment, and proposed that enterprises need to focus on five layers: compute, models, data & knowledge, agentic platform, and applications. He also noted that token costs are often high because too much useless information is fed to the model.

Headroom is an open-source context compression layer that reduces token consumption by 50-90% by compressing all content (tool outputs, logs, RAG chunks, files, conversation history) before it reaches the LLM. It offers multiple integration modes (library, proxy, agent wrap, MCP server), supports various AI agents (Claude Code, Codex, Cursor, etc.), and preserves answer accuracy on benchmarks. The community has saved over 60B tokens.

Fluiq is an AI Ops platform covering security, optimization, observability, and evaluations. It offers a free tier for early signups and reviews, and detects threats with minimal code.

At the 2026 China AIGC Industry Summit, Wang Xiaoye, Technical Director of Amazon Web Services, pointed out that 87% of enterprises claim to have deployed AI at scale, but only 10% have gained real production value. He emphasized that enterprise-grade Agent deployment must bridge four major gaps: model selection, construction complexity, usage threshold, and talent shortage. He introduced AWS's five-layer architecture—compute, model, data, harness platform, and agent applications—and products like Quick to help enterprises move from demo to production.

A new tool allows comparing prices for 385 AI models across 9 providers, helping users find the cheapest option.

A high-speed LLM playground achieving 3000 tokens per second, featuring an open web UI.

Anthropic has released Claude Opus 4.8, an upgrade to Opus 4.7 with improvements in coding, agent work, reasoning, and knowledge work. New features include effort control, dynamic workflows, and live Messages API updates. Pricing remains unchanged at $5/$25 per million tokens for standard and $10/$50 for fast mode (2.5x speed). Early testers report cost parity with GPT-5.5 and fewer tool steps. The company also outlined its roadmap including Mythos-class models and Project Glasswing for cybersecurity.

Aryabhata 2 is a reasoning-focused language model for competitive STEM exams like JEE and NEET, fine-tuned via reinforcement learning on GPT-OSS-20B using PhysicsWallah's question banks. It achieves up to 64% fewer output tokens while outperforming the base model on multiple benchmarks.

This paper presents RightNow-Arabic-0.5B-Turbo, a 518M-parameter Arabic-specialized LLM built on Qwen2.5-0.5B using vocabulary injection and edge-first deployment. It achieves 35.9% mean accuracy on Arabic benchmarks, outperforming all same-class open models, and ties Falcon-H1-1.5B on COPA-ar at one-third the size. The quantized model is 398 MB and delivers 635 tokens/s on a single H100, enabling efficient edge deployment.

This paper proposes COM, a strategy that integrates geometric constraints into token initialization and training to preserve the inherent continuity and ordinality of time series tokens, consistently improving the performance of token-based time series LLMs on multiple benchmarks.

Release of llm-anthropic 0.25.1 adds support for Claude Opus 4.8, fast mode option for eligible accounts, and changes default max_tokens to each model's maximum output.

This article demonstrates how to implement a context pruning pipeline for long-running AI agents to manage conversational memory efficiently using semantic similarity. It covers using sentence transformer embedding models, computing similarities, and assembling a pruned context window.

Open Agent Tools (oats) is a self-hosted AI framework that enables small-to-large local models to use local source code for tool-calling, freeing up expensive large model tokens by delegating tasks to smaller models.

ICG is a novel framework that integrates MLLM-based prompting with personalized preference alignment to generate high-quality, contextually relevant cover images. It extracts semantic features via meta tokens, refines them with user embeddings, and injects personalized context into diffusion models. A multi-reward learning strategy combines public rewards with a personalized preference model, eliminating the need for labeled supervision. Experiments show improvements in image quality, semantic fidelity, and personalization, boosting user appeal and recommendation accuracy.

At Databricks, we’ve built a unique inference platform that serves every frontier model, from open source to proprietary, powering some of the largest agentic applications. Serving over 120T tokens per month, we tackle challenges of reliability and latency through abstractions like model units for capacity management, cost-aware load balancing and autoscaling that save over 80% GPU costs, and runtime reliability mechanisms including black-box health checks that detect silent failures. Profiling multimodal bottlenecks unlocked 3x throughput gains.

Mr. Guy Invests is a free, beginner-friendly stock research and portfolio tracker that leverages public SEC filings to track hedge fund and insider activity, offers an AI stock tutor, a $100K virtual trading challenge, daily market briefs, and more. Free tier has daily limits; Pro is $4.99/month for unlimited access.

One month after DeepSeek V4's release, the open-source community unveiled Reasonix, a tool specifically designed to minimize API costs by maximizing cache efficiency. It achieves a staggering 99.82% cache hit rate, reducing a $61 bill for 400M+ tokens to just $12.

This study challenges the assumption that high benchmark scores reflect true visual understanding in vision-language models (VLMs). By removing a large fraction of image tokens with minimal performance drop, the authors reveal a mismatch between accuracy and visual grounding. Through multi-level analyses including global degradation, localized occlusion, question reformulation, answer-space expansion, decision-level analysis, and layer-wise vision-token geometry, they find that models are less sensitive to fine-grained visual evidence than expected, and that visual tokens become more similar in deeper layers. The results indicate that current benchmarks are insufficient for evaluating fine-grained visual grounding in VLMs.

NVIDIA's Gated DeltaNet-2 is a linear attention layer that decouples memory erasing and writing into channel-wise gates. Trained at 1.3B parameters on 100B FineWeb-Edu tokens, it outperforms Mamba-2, Gated DeltaNet, KDA, and Mamba-3 in language modeling, commonsense reasoning, and long-context retrieval, with the largest gains on RULER benchmarks.

Deepseek is making the 75 percent discount on its top model V4-Pro permanent. At $0.435 per million input tokens, it's at least 11.5 times cheaper than GPT-5.5 and over 34 times cheaper on output. For token-hungry agentic systems, this kind of pricing could squeeze Western providers hard.

DeepSeek announced permanent price cuts for its V4-Pro API. Meanwhile, CATL, JD, and NetEase are in talks to invest in DeepSeek's first external funding round. Founder Liang Wenfeng emphasizes prioritizing AGI research and maintaining open-source principles.

NVIDIA introduces Nemotron-Labs Diffusion language models that achieve up to 6.4x faster inference than autoregressive models while maintaining high accuracy by generating tokens in parallel and refining them iteratively. The models support three modes: autoregressive, diffusion, and self-speculation. The 8B model outperforms Qwen3 8B by 1.2% accuracy.

This paper proposes the Ablate-to-Validate diagnostic principle and its instantiation, the Token Replacement Test (TRT), to determine whether vision-language models (VLMs) genuinely use continuous latent tokens for reasoning. Experiments show that VLMs retain most performance gains even when token content is corrupted or replaced, indicating that accuracy improvements are a misleading proxy for latent-token reasoning.

Alibaba's Qwen team announced Qwen3.7-Max, their most advanced agent model, featuring a 1M-token context window, extended-thinking mode, and strong benchmark scores (56.6 on AI Index, 5th overall). The model excels at coding, debugging, and long-horizon autonomous tasks but has caveats like reduced factual recall on AA-Omniscience and no independent verification of long-context reliability.