AI Model Weekly Roundup - May 4–10, 2026
This week in AI: Mistral Medium 3.5 ships as a 128B open-weight flagship, Grok 4.3 stakes its 1M-context claim, and GPT-5.5 Instant undercuts the mid-tier.
Read more →LMRank Blog
AI model news, analysis, pricing context, and release notes.
AI Model Weekly Roundup - May 18–24, 2026
This week in AI: DeepSeek's reasonix coding agent tops Hacker News, Google ships Gemini 3.5 Flash, xAI launches Grok Build 0.1, and memory costs climb.
Read more →AI Model Weekly Roundup - May 25–31, 2026
This week in AI: DeepSWE crowns GPT-5.5 and flags Claude Opus 4.8, GitHub Copilot adds a 57x GPT-5.5 multiplier, and distillation drama hits Opus 4.8.
Read more →Creative Benchmarks: Testing What AI Models Can Build
LMRank's Creative HTML Generation benchmarks test what AI models can actually build - landing pages, 3D simulations, and full projects from a single prompt.
Read more →DeepSeek Makes Its 75% API Discount Permanent
DeepSeek confirmed its V4 Pro 75% API discount won't expire. At $0.50 per million tokens with a 9.0 LMRank score, this changes the economics of frontier AI.
Read more →DeepSeek V4-Flash: Open-Weight, Million-Token Context
DeepSeek V4-Flash brings a 1M-token context window to open-weight AI: 284B params, 13B active, MIT license, and coding scores that rival closed frontier models.
Read more →GPT-5.5's Hidden Cost Surge: What OpenRouter Found
OpenRouter's analysis reveals GPT-5.5 costs 49-92% more in practice than earlier GPT-5 models. Here's the breakdown by prompt length and the best alternatives.
Read more →Grok 4.3 vs Gemini Ultra 2: The Million-Token Showdown
xAI's Grok 4.3 undercuts Gemini Ultra 2 by 90% on output pricing while matching its 1M token context window. Here's the breakdown.
Read more →Kimi K2.6 vs Qwen3.6 35B: Open-Weight Coding Crown
Kimi K2.6 vs Qwen3.6 35B A3B: two open-weight coding models compared on price, LMRank score, and agentic reliability. The cheaper Qwen edges ahead on value.
Read more →MiniMax M3: Open-Weight Coding, 1M Context, Multimodal
MiniMax M3 ships open-weights with 1M context, native multimodal, and frontier coding claims at $0.30/M input tokens. Here's why it matters.
Read more →Mistral Medium 3.5: 128B Open-Weight Agentic Coder
Mistral's new 128B dense model packs a 256K context window, open weights, and claims 77.6% on SWE-Bench. Here's why the medium tier just got interesting.
Read more →Qwen3.7 Max: The $2.50 Sleeper Flagship
Qwen3.7 Max pairs a 9.0 LMRank score and 1M-token context with $2.50/M input pricing - one of the best-value frontier models you can call today.
Read more →State of Large Language Models - May 2026
The state of large language models in May 2026: who leads, who's overpriced, and which model to actually use - with live benchmark and pricing data from LMRank.
Read more →When Claude Builds Claude: Inside Anthropic's Recursion
Anthropic says Claude now writes 80% of its own code and ships 8x faster - then called for a global AI pause. Inside the recursive self-improvement data.
Read more →AI Model Weekly Roundup - June 2–8, 2026
This week in AI: Anthropic scales Claude Mythos into critical infrastructure, Google ships on-device Gemma 4 QAT, Meta delays again, and Linux wants Claude.
Read more →Mixtral 8x22B v2: 512K Context, 40% Price Cut
Mistral ships Mixtral 8x22B v2 with a 512K context window and 40% price cut, undercutting every long-context competitor on the market.
Read more →The US Government Shut Down Claude Fable 5 in 3 Days
Anthropic launched Claude Fable 5 on June 9; a US export-control directive killed it worldwide by June 12 - the first government shutdown of a live AI API.
Read more →The AI Price War of 2026: The Race to Zero
The 2026 AI price war is here: DeepSeek cut V4.1 15%, OpenAI added 90% cache discounts, and Google hit $0.10/M tokens. What falling prices mean for developers.
Read more →AI Model Weekly Roundup - June 9–15, 2026
This week in AI: Claude Fable 5 was killed by a US export directive 72 hours after launch, Mistral shipped 512K-context Mixtral 8x22B v2, and prices fell.
Read more →Kimi K2.7 Code: 1T Params, Self-Made Benchmarks
Moonshot AI's Kimi K2.7 Code packs 1T parameters and a $0.95/M-token price, but every launch benchmark is proprietary. Here's why that matters.
Read more →Model Routers: OpenRouter Fusion vs Sakana Fugu Ultra
Model routers are here: OpenRouter Fusion's parallel synthesis vs Sakana AI's learned Fugu Ultra orchestration - how they compare on latency, control, and cost.
Read more →Google's 38% Gemini AI Plus Cut Fuels the Price War
Google slashed its AI Plus subscription from $7.99 to $4.99 per month while doubling storage. The consumer AI price war just got real.
Read more →GPT-5.6 Sol/Terra/Luna: OpenAI's Gated Release
OpenAI's GPT-5.6 Sol/Terra/Luna shipped only after US government preclearance - a tiered, gated release that could set the template for frontier AI launches.
Read more →Weekly AI Model Roundup - June 22–29, 2026
This week in AI: OpenAI launches GPT-5.6 Sol under government gating, Claude Fable 5 stays offline, and Sakana Fugu Ultra hits frontier scores via routing.
Read more →LongCat-2.0: China's 1.6T-Parameter Coding Model Trained Without a Single Nvidia GPU
Meituan's LongCat-2.0 is a 1.6T-parameter MoE coding model trained entirely on domestic Chinese ASICs. It challenges the assumption that frontier AI requires Nvidia hardware.
Read more →DeepSeek V4 Pro vs GPT-5.5 vs Claude Opus 4.8: Agentic Coding Price-to-Performance in 2026
DeepSeek V4 Pro, GPT-5.5, and Claude Opus 4.8 compared head-to-head for agentic coding. Price, quality, and real engineering trade-offs for mid-2026 teams.
Read more →Agentic Coding Showdown: DeepSeek V4 Pro vs GPT-5.5 vs Opus 4.8
DeepSeek V4 Pro, GPT-5.5, and Claude Opus 4.8 compared for agentic coding in mid-2026. Pricing, benchmarks, and which model to pick for real engineering work.
Read more →AI Model Roundup: June 30–July 6, 2026
Claude Sonnet 5 launches with hidden cost increases. OpenAI previews GPT-5.6 Sol/Terra/Luna. Meituan open-sources LongCat 2.0. Plus new diffusion models and benchmark shifts.
Read more →Tencent Hy3 vs DeepSeek V4 Pro: The New Open-Weight Price Fight
Tencent Hy3 enters LMRank near DeepSeek V4 Pro with dramatically lower input pricing, strong agentic benchmarks, Apache 2.0 weights, and a clear role as a cheaper production worker model.
Read more →DeepSeek Peak Pricing: The End of the AI Token Free Lunch
DeepSeek's shift from price cuts to peak-hour surge pricing reveals capacity constraints. What this means for inference economics and the 'race to zero.'
Read more →Metered Billing Ends the Flat-Rate AI Era
Anthropic, OpenAI, and Meta have all shifted flagship models and agents to metered billing. Users must adopt token-cost discipline or pay more.
Read more →
GPT-5.6 Sol, Terra, Luna: Compared, and the One Rule
OpenAI shipped GPT-5.6 as three tiers with a reasoning-effort dial. The benchmark data says it is really three models at max effort. Here is the comparison and the rule of thumb.
Read more →Thinking Machines Inkling: Open Weights, Not Just Another Leaderboard Model
Thinking Machines Lab's first model, Inkling, trades top-line leaderboard performance for something harder to buy: an open multimodal base that developers can fine-tune and reshape.
Read more →Kimi K3 Pricing Ends the 'Open Weights = Cheap' Era
Moonshot AI's Kimi K3 launch at $3/$15 per MTok — matching Anthropic's Sonnet tier — shatters the assumption that open-weight models are budget options.
Read more →Kimi K3 and China's AI Offensive: Weekly Roundup July 13-19
Moonshot AI's Kimi K3 launch, Alibaba's Qwen3.8 Max Preview, and Thinking Machines Inkling signal a decisive shift in open-weight AI leadership during WAIC 2026.
Read more →AI Model Roundup: Aug 17-23, 2026
Open-weight releases from Z.AI, Ornith AI, and Alibaba compete on coding benchmarks this week. Pricing shifts from OpenAI and DeepSeek change the cost landscape.
Read more →The Hallucination Trade-Off: Agent Reliability Replaces Benchmark Chasing in 2026
Chinese labs decouple quality from hallucination rates as Grok 4.6, Qwen3.8-Flash-Next, and GLM-5.3-Flash shift the war from cheapest model to most reliable agent infrastructure.
Read more →