https://benchr.org/ Pricing, benchmarks, and use-case fit for AI model selection. en-us Fri, 12 Jun 2026 11:18:15 +0000 corrections@benchr.org (benchr) https://benchr.org/deprecations/openai-october-2026-retirements https://benchr.org/deprecations/openai-october-2026-retirements Fri, 12 Jun 2026 00:00:00 +0000 On October 23, 2026, OpenAI retires eleven model IDs at once: GPT-4o, GPT-4, GPT-4 Turbo, GPT-3.5 Turbo, GPT-4.1 nano, GPT Image 1, o1, o1-pro, o3-mini, and o4-mini. The Assistants API goes August 26. Every date, replacement, and audit step. https://benchr.org/deprecations/gpt-4o https://benchr.org/deprecations/gpt-4o Fri, 12 Jun 2026 00:00:00 +0000 OpenAI retires the gpt-4o-2024-05-13 snapshot on October 23, 2026. The official replacement is GPT-5.5 at $5/$30 — but GPT-5 at $1.25/$10 is the better cost match for most GPT-4o workloads. The full decision. https://benchr.org/deprecations/gemini-2-5-pro https://benchr.org/deprecations/gemini-2-5-pro Fri, 12 Jun 2026 00:00:00 +0000 Google retires gemini-2.5-pro and gemini-2.5-flash on October 16, 2026. The catch: Gemini 3.5 Flash bills $1.50/1M input vs 2.5 Flash's $0.30 — a 5× jump. Here's the migration math and the cheaper detours. https://benchr.org/contact https://benchr.org/contact Fri, 12 Jun 2026 00:00:00 +0000 How to reach benchr for general questions, corrections, and security reports. https://benchr.org/deprecations/claude-sonnet-4 https://benchr.org/deprecations/claude-sonnet-4 Fri, 12 Jun 2026 00:00:00 +0000 Anthropic retires claude-sonnet-4-20250514 on June 15, 2026. The replacement, Claude Sonnet 4.6, costs the same $3/$15, scores 79.6% on SWE-bench vs 72.7%, and carries a 1M context window. Here's the swap, step by step. https://benchr.org/deprecations/claude-opus-4 https://benchr.org/deprecations/claude-opus-4 Fri, 12 Jun 2026 00:00:00 +0000 Claude Opus 4 retires June 15, 2026 and Opus 4.1 follows August 5. Migrating to Opus 4.8 cuts the bill from $15/$75 to $5/$25 per million tokens and lifts SWE-bench from 72.5% to 88.6%. One API parameter will bite you. https://benchr.org/deprecations https://benchr.org/deprecations Fri, 12 Jun 2026 00:00:00 +0000 Announced AI model retirements across Anthropic, OpenAI, and Google, with shutdown dates, official replacements, and migration cost analysis. Verified against official provider deprecation documentation. https://benchr.org/errors https://benchr.org/errors Fri, 12 Jun 2026 00:00:00 +0000 Common AI API errors across OpenAI, Anthropic, and Google Gemini, each verified against official provider documentation, with causes, fixes, and migration alternatives. https://benchr.org/articles/gpt-5-4-review https://benchr.org/articles/gpt-5-4-review Wed, 10 Jun 2026 00:00:00 +0000 GPT-5.4 review, three months after launch: $2.50/$15 pricing, 1M context, 75% OSWorld computer use, and why losing the flagship crown to GPT-5.5 made it the value pick. https://benchr.org/pricing/gpt-5-4 https://benchr.org/pricing/gpt-5-4 Wed, 10 Jun 2026 00:00:00 +0000 GPT-5.4 costs $2.50/1M input and $15/1M output — double GPT-5, half of GPT-5.5. GPT-5.4's $0.25 caching, the 272K long-context surcharge, and when the middle tier is the value pick. https://benchr.org/ar/articles/claude-fable-5-launch https://benchr.org/ar/articles/claude-fable-5-launch Wed, 10 Jun 2026 00:00:00 +0000 صدر Claude Fable 5 في 9 يونيو 2026 بسعر $10/$50 لكل مليون رمز. إنه Mythos مع مصنفات أمان: ما الذي يحجبه اللجام، وما الذي تكسبه فوق Opus 4.8، ونافذة الأسبوعين المجانية. https://benchr.org/articles/claude-fable-5-launch https://benchr.org/articles/claude-fable-5-launch Wed, 10 Jun 2026 00:00:00 +0000 Claude Fable 5 shipped June 9, 2026 at $10/$50 per million tokens. It's Mythos with safety classifiers: what the leash blocks, what you get over Opus 4.8, and the two-week free window. https://benchr.org/pricing/claude-fable-5 https://benchr.org/pricing/claude-fable-5 Wed, 10 Jun 2026 00:00:00 +0000 Claude Fable 5 costs $10/1M input and $50/1M output — double Opus 4.8. Caching math, the tokenizer surcharge nobody mentions, fallback economics, and the free window through June 22, 2026. https://benchr.org/articles/anthropic-glasswing-claude-security https://benchr.org/articles/anthropic-glasswing-claude-security Tue, 09 Jun 2026 00:00:00 +0000 Anthropic extended Project Glasswing to about 150 more organizations and launched Claude Security, a product that uses Claude Opus 4.8 to scan codebases and suggest patches. What it means for developers. https://benchr.org/articles/uk-google-ai-search-publishers https://benchr.org/articles/uk-google-ai-search-publishers Tue, 09 Jun 2026 00:00:00 +0000 The UK's CMA now requires Google to let publishers opt out of having their content power AI features in Search — without losing standard rankings. A world first, and a template other regulators may copy. What it means. https://benchr.org/articles/grok-4-3-default-and-pricing https://benchr.org/articles/grok-4-3-default-and-pricing Tue, 09 Jun 2026 00:00:00 +0000 xAI made Grok 4.3 the default model and redirected deprecated text slugs to it — billed at Grok 4.3 prices. If your app still calls Grok 4.1 Fast or Grok 3, your costs may have jumped. Here's how to check. https://benchr.org/articles/gpt-5-5-pricing-explained https://benchr.org/articles/gpt-5-5-pricing-explained Tue, 09 Jun 2026 00:00:00 +0000 GPT-5.5 costs $5/$30 per million tokens — until a prompt crosses 272K input tokens, when the whole session reprices higher. And GPT-5.5 Pro is a different model at $30/$180. What both mean for your bill. https://benchr.org/articles/webmcp-agent-readable-websites https://benchr.org/articles/webmcp-agent-readable-websites Tue, 09 Jun 2026 00:00:00 +0000 WebMCP is a proposed web standard, backed by Google and Microsoft, that lets sites expose structured tools to AI agents instead of making them guess from the DOM. What it is, where it stands, and why site owners should care. https://benchr.org/articles/google-ai-search-agent-layer https://benchr.org/articles/google-ai-search-agent-layer Tue, 09 Jun 2026 00:00:00 +0000 Google's AI Mode is shifting from summarizing results to acting on them with background agents. Meanwhile AI Overviews keep cutting clicks. The real question for site owners: which pages survive? https://benchr.org/articles/google-ai-agents-io-2026 https://benchr.org/articles/google-ai-agents-io-2026 Tue, 09 Jun 2026 00:00:00 +0000 At I/O 2026 Google shipped Gemini 3.5 Flash as an agent-first model, started moving Gemini CLI users to Antigravity by June 18, and reframed developer AI around autonomous agents. What changes for builders. https://benchr.org/articles/claude-opus-4-8-launch https://benchr.org/articles/claude-opus-4-8-launch Tue, 09 Jun 2026 00:00:00 +0000 Claude Opus 4.8 shipped May 28, 2026 at the same $5/$25 pricing as 4.7. The real story is agentic coding, dynamic workflows, and a Mythos-class model heading to all customers. What's official and what to make of it. https://benchr.org/pricing/qwen-3-6-27b https://benchr.org/pricing/qwen-3-6-27b Sat, 06 Jun 2026 00:00:00 +0000 Qwen 3.6-27B is a free open-weight model from Alibaba. Via API ~$0.20/1M input. Strong multilingual (Arabic, Chinese, European), 27B MoE, self-hostable on consumer GPUs. https://benchr.org/pricing/phi-4 https://benchr.org/pricing/phi-4 Sat, 06 Jun 2026 00:00:00 +0000 Phi-4 is free to self-host from Microsoft. Via API ~$0.07/1M input. 14B parameters, MIT license, runs on a laptop GPU. Strong reasoning benchmarks for its size class. https://benchr.org/articles/openai-api-pricing-guide https://benchr.org/articles/openai-api-pricing-guide Sat, 06 Jun 2026 00:00:00 +0000 Complete guide to OpenAI API pricing in 2026. Review input, output, prompt caching, and batch execution pricing for all GPT models. Sourced from official docs. https://benchr.org/pricing/mistral-medium-3-5 https://benchr.org/pricing/mistral-medium-3-5 Sat, 06 Jun 2026 00:00:00 +0000 Mistral Medium 3.5 costs $1.50/1M input and $7.50/1M output. Vision and PDF support, 128K context, EU-resident infrastructure. Mistral's multimodal production model. https://benchr.org/pricing/mistral-large-3 https://benchr.org/pricing/mistral-large-3 Sat, 06 Jun 2026 00:00:00 +0000 Mistral Large 3 costs $0.50/1M input and $1.50/1M output. Apache 2.0 license for self-hosting, EU data residency available. Strong multilingual and code performance at a low price. https://benchr.org/pricing/llama-4-scout https://benchr.org/pricing/llama-4-scout Sat, 06 Jun 2026 00:00:00 +0000 Llama 4 Scout is free to self-host from Meta. Via API ~$0.11/1M input. 10 million token context — the largest available window. Lightweight MoE for extreme long-context tasks. https://benchr.org/pricing/llama-4-maverick https://benchr.org/pricing/llama-4-maverick Sat, 06 Jun 2026 00:00:00 +0000 Llama 4 Maverick is free to self-host from Meta. Via API ~$0.20/1M input. Multimodal (vision), 1M context, MoE architecture. The open-source choice for multimodal pipelines. https://benchr.org/pricing/kimi-k2-6 https://benchr.org/pricing/kimi-k2-6 Sat, 06 Jun 2026 00:00:00 +0000 Kimi K2.6 costs $0.95/1M input and $4/1M output. 80.2% SWE-bench, 90.5% GPQA Diamond, 200K context. Moonshot AI's model combining coding and science strengths under $1. https://benchr.org/pricing/grok-4-3 https://benchr.org/pricing/grok-4-3 Sat, 06 Jun 2026 00:00:00 +0000 Grok 4.3 costs $1.25/1M input and $2.50/1M output. Native real-time web search, 256K context, xAI's production model. Unusually low output price relative to input. https://benchr.org/compare/gpt-5-vs-gemini-3-5-flash https://benchr.org/compare/gpt-5-vs-gemini-3-5-flash Sat, 06 Jun 2026 00:00:00 +0000 GPT-5 ($1.25/$10) vs Gemini 3.5 Flash ($1.50/$9): nearly the same price, completely different machines. Speed, context, free tier, and output ceilings compared. https://benchr.org/pricing/gpt-5-mini https://benchr.org/pricing/gpt-5-mini Sat, 06 Jun 2026 00:00:00 +0000 GPT-5 Mini costs $0.25/1M input and $2/1M output. 160 tok/s, 48% SWE-bench, 128K context. OpenAI's cheapest API model for classification, routing, and volume pipelines. https://benchr.org/pricing/gemini-3-1-pro https://benchr.org/pricing/gemini-3-1-pro Sat, 06 Jun 2026 00:00:00 +0000 Gemini 3.1 Pro costs $2/1M input and $12/1M output. 1M context window, 94.3% GPQA Diamond, native multimodal. Google's frontier model for long-document and research workloads. https://benchr.org/compare/deepseek-v4-pro-vs-gpt-5 https://benchr.org/compare/deepseek-v4-pro-vs-gpt-5 Sat, 06 Jun 2026 00:00:00 +0000 DeepSeek V4-Pro vs GPT-5: 11.5× cheaper output ($0.87 vs $10) and a higher SWE-bench score (80.6% vs 74.9%). Where the catch really is, with monthly cost math. https://benchr.org/compare/deepseek-v4-pro-vs-claude-sonnet-4-6 https://benchr.org/compare/deepseek-v4-pro-vs-claude-sonnet-4-6 Sat, 06 Jun 2026 00:00:00 +0000 DeepSeek V4-Pro ($0.435/$0.87) vs Claude Sonnet 4.6 ($3/$15): one SWE-bench point apart, a 10× price gap. What the cheaper model gives up, with real pipeline math. https://benchr.org/pricing/deepseek-v4-flash https://benchr.org/pricing/deepseek-v4-flash Sat, 06 Jun 2026 00:00:00 +0000 DeepSeek V4-Flash costs $0.14/1M input and $0.28/1M output. 79% SWE-bench, 1M context, MIT license. The cheapest model with near-Sonnet coding performance. https://benchr.org/compare/claude-sonnet-4-6-vs-gpt-5 https://benchr.org/compare/claude-sonnet-4-6-vs-gpt-5 Sat, 06 Jun 2026 00:00:00 +0000 Claude Sonnet 4.6 ($3/$15) vs GPT-5 ($1.25/$10): SWE-bench scores, context windows, caching math, and a concrete monthly cost example for the daily-driver decision. https://benchr.org/pricing/claude-sonnet-4-6 https://benchr.org/pricing/claude-sonnet-4-6 Sat, 06 Jun 2026 00:00:00 +0000 Claude Sonnet 4.6 costs $3/1M input and $15/1M output. 79.6% SWE-bench, 200K context, 64K max output. The balanced choice between Haiku's speed and Opus's depth. https://benchr.org/pricing/claude-opus-4-7 https://benchr.org/pricing/claude-opus-4-7 Sat, 06 Jun 2026 00:00:00 +0000 Claude Opus 4.7 costs $5/1M input and $25/1M output. Same price as Opus 4.8, one point lower on SWE-bench. When to use 4.7 vs 4.8 vs Sonnet 4.6. https://benchr.org/pricing/claude-haiku-4-5 https://benchr.org/pricing/claude-haiku-4-5 Sat, 06 Jun 2026 00:00:00 +0000 Claude Haiku 4.5 costs $1/1M input and $5/1M output. 145 tok/s, 73.3% SWE-bench, 200K context. Anthropic's speed and volume tier for high-throughput pipelines.