All notable changes to llm-prices will be documented here.

• Kimi K2.5 (kimi-k2.5): $0.40/$1.90, 262k ctx; Moonshot AI — Jul 2026 successor to K2, cheaper than K2.6
• Solar Pro 3 (solar-pro-3): $0.15/$0.60, 128k ctx; Upstage (new #45) — Korean AI enterprise model
• OLMo 3 32B Think (olmo-3-32b-think): $0.15/$0.50, 65k ctx; AllenAI (new #46) — fully open Apache 2.0 reasoning model
• INTELLECT-3 (intellect-3): $0.20/$1.10, 131k ctx; Prime Intellect (new #47) — distributed training reasoning model

• Models: 273 → 277 | Providers: 44 → 47

• llm-prices new command: shows recently added models grouped by release version. Supports --version v0.1.42 to filter and --json for structured output. Useful for tracking the rapidly evolving LLM market.

• Models: 273 (unchanged) | Providers: 44 (unchanged) | Commands: 6 → 7

• Mistral Medium 3.5 (mistral-medium-3-5): $1.5/$7.5, 262k ctx; dense 128B updated from Medium 3
• Qwen3.6 Flash (qwen3.6-flash): $0.1875/$1.125, 1M ctx; fast multimodal (text+image+video+audio)
• Qwen3.6 35B MoE (qwen3.6-35b-moe): $0.15/$1.0, 262k ctx; 35B total / 3B active params
• Qwen3.6 27B (qwen3.6-27b): $0.32/$3.2, 262k ctx; dense 27B multimodal
• Qwen3.6 Max Preview (qwen3.6-max-preview): $1.04/$6.24, 262k ctx; sparse MoE frontier preview
• Qwen3.5 Plus v2 (qwen3.5-plus-v2): $0.30/$1.80, 1M ctx; Apr 2026 update to qwen3.5-plus
• DeepSeek V4 Pro (deepseek-v4-pro): $0.435/$0.87, 1M ctx; 1.6T-param MoE flagship
• DeepSeek V4 Flash (deepseek-v4-flash): $0.112/$0.224, 1M ctx; 284B-param efficient MoE
• Moonshot Kimi K2.6 (kimi-k2.6): $0.73/$3.49, 262k ctx; next-gen multimodal from Moonshot AI
• Perceptron Mk1 (perceptron-mk1): $0.15/$1.5, 32k ctx; new provider — video+embodied VLM

• Perceptron (#44): Perceptron AI, specializing in video understanding and embodied AI reasoning

• Models: 267 → 273 | Providers: 43 → 44

• OpenAI GPT-5.1-Codex-Mini (gpt-5.1-codex-mini): $0.25/$2.00, 400k ctx; cheap coding specialist model
• OpenAI GPT-5.2 Pro (gpt-5.2-pro): $21.00/$168.00, 400k ctx; ultra-premium agentic/coding tier (most expensive GPT-5 variant)
• OpenAI GPT-OSS Safeguard 20B (gpt-oss-safeguard-20b): $0.075/$0.30, 131k ctx; open-weights safety reasoning model built on gpt-oss-20b
• Google Gemini 3 Pro Image (gemini-3-pro-image): $2.00/$12.00, 65k ctx; image generation and editing model built on Gemini 3 Pro
• Google Gemini 3.1 Flash Image (gemini-3.1-flash-image): $0.50/$3.00, 131k ctx; cheaper image generation tier

• gpt-5-codex context window updated 272k → 400k (matches gpt-5.1-codex spec)

• Models: 262 → 267 | Providers: 43 (unchanged)

• Qwen3 VL series (new multimodal generation): qwen3-vl-8b ($0.08/$0.50, 256k), qwen3-vl-30b-moe ($0.13/$0.52, 262k), qwen3-vl-32b ($0.104/$0.416, 262k), qwen3-vl-235b-moe ($0.20/$0.88, 262k); Qwen3-generation vision-language models
• Qwen3.5 Flash (qwen3.5-flash): $0.065/$0.26, 1M ctx; ultra-cheap flash tier
• Qwen3 Next 80B MoE (qwen3-next-80b-moe): $0.09/$1.10, 262k ctx; next-gen reasoning MoE
• Anthropic Claude Opus 4.7 Fast (claude-opus-4.7-fast): $30/$150, 1M ctx; high-throughput priority-capacity tier (6× standard price)

• Models: 255 → 262 | Providers: 43 (unchanged)

• Xiaomi (new provider #36): mimo-v2-flash ($0.10/$0.30, 262k ctx), mimo-v2.5 ($0.40/$2.00, 1M ctx), mimo-v2.5-pro ($1.00/$3.00, 1M ctx); MiMo series beats GPT-4o on math benchmarks
• Z.AI (new provider #37): glm-4.7-flash ($0.06/$0.40, 202k ctx), glm-4.7 ($0.40/$1.75), glm-5 ($0.60/$1.92), glm-5-turbo ($1.20/$4.00); GLM 5.x successor series
• Tencent (new provider #38): hunyuan-a13b ($0.14/$0.57, 131k ctx, 13B active MoE), hunyuan-hy3 ($0.066/$0.26, 262k ctx)
• Arcee AI (new provider #39): arcee-trinity-mini ($0.045/$0.15), arcee-trinity-large ($0.22/$0.85, 262k ctx), arcee-maestro ($0.90/$3.30)
• InclusionAI (new provider #40): ling-2.6-flash ($0.01/$0.03, 262k ctx — one of cheapest available), ling-2.6-1t ($0.30/$2.50, 1T param), ring-2.6-1t ($0.075/$0.625, 1T thinking model)
• StepFun (new provider #41): step-3.5-flash ($0.10/$0.30, 262k ctx); Chinese AI lab
• Reka (new provider #42): reka-flash-3 ($0.10/$0.20, 65k ctx); multimodal research lab
• Morph (new provider #43): morph-v3-fast ($0.80/$1.20, 81k ctx), morph-v3-large ($0.90/$1.90, 262k ctx); specialized code editing/apply models

• Models: 238 → 255 | Providers: 35 → 43

• OpenAI new models: gpt-5.4-pro ($30/$180, 1M ctx flagship), gpt-5.5-pro ($30/$180, 1M ctx latest flagship)
• xAI new models: grok-4.20 ($1.25/$2.50, 2M ctx standard), grok-4.20-agent ($2/$6, 2M ctx multi-agent mode)
• Qwen new models: qwen3.5-plus ($0.26/$1.56, 1M ctx Feb 2026), qwen3.5-plus-v2 ($0.30/$1.80, 1M ctx Apr 2026)

• Models: 232 → 238 | Providers: 35 (unchanged)

• Qwen new models: qwen3.5-27b (dense 27B, $0.195/$1.56), qwen3.5-35b (A3B MoE, $0.14/$1.00), qwen3.5-122b (A10B MoE, $0.26/$2.08), qwen3.6-35b (MoE, $0.15/$1.00), qwen3.6-plus (1M ctx, $0.325/$1.95), qwen3.6-max (flagship preview, $1.04/$6.24), qwen-vl-72b (VL multimodal, $0.25/$0.75)
• Mistral new model: mistral-small-24b-2501 — Jan 2025 update, ultra-cheap $0.05/$0.08, 32k ctx

• Models: 224 → 232 | Providers: 35 (unchanged)

• Liquid AI (new provider #33): lfm-2-24b — $0.03/$0.12 per Mtok; non-transformer liquid architecture; ultra-cheap inference
• Inception AI (new provider #34): mercury-2 — $0.25/$0.75 per Mtok, 128k ctx; diffusion-based LLM architecture
• Writer (new provider #35): palmyra-x5 — $0.60/$6.00 per Mtok, 1M+ ctx; enterprise writing workflows
• MiniMax new models: minimax-m1 (1M ctx, $0.40/$2.20), minimax-m2.1 (200k ctx, $0.29/$0.95)
• Qwen new models: qwen3.5-397b (397B flagship, $0.39/$2.34), qwen3.6-27b (new gen dense), qwen3.6-flash (1M ctx)
• Google new model: gemma-3n-e4b — new Gemma-3N mobile-optimized architecture
• Mistral new model: ministral-8b-2512 — Dec 2025 update, 262k ctx, flat $0.15/Mtok

• Models: 214 → 224 | Providers: 32 → 35

• Baidu (new provider #32): ERNIE 4.5 21B MoE, ERNIE 4.5 300B MoE, ERNIE 4.5 VL 28B, ERNIE 4.5 VL 424B — China's leading conversational AI platform with 300M+ users

• Models: 210 → 214 | Providers: 31 → 32

• Gemini 3.5 Flash (gemini-3.5-flash): $0.75/$4.50 per Mtok, 1M ctx
  • Launched at Google I/O 2026 — "frontier intelligence with speed"
  • Includes thinking tokens in output price (reasoning-capable)
  • Context: 1,048,576 tokens
  • Source: https://ai.google.dev/pricing
• Total: 209 → 210 models (same 31 providers)

• New provider: Microsoft (2 models):
  • Phi-4 14B (phi-4): $0.07/$0.14, 16k ctx — state-of-the-art SLM; strong reasoning
  • Phi-4 Mini (phi-4-mini): $0.08/$0.35, 131k ctx — efficient on-device capable model
• New provider: ZhipuAI (4 models — GLM family):
  • GLM-4 32B (glm-4-32b): $0.10/$0.10, 128k ctx — flat in=out; strong multilingual
  • GLM-4.5 Air (glm-4.5-air): $0.13/$0.85, 131k ctx — mid-tier speed/quality
  • GLM-4.7 Flash (glm-4.7-flash): $0.06/$0.40, 202k ctx — cheapest GLM; large context
  • GLM-4.7 (glm-4.7): $0.40/$1.75, 202k ctx — flagship GLM model
• New provider: MiniMax (3 models):
  • MiniMax-01 (minimax-01): $0.20/$1.10, 1M ctx — 1 million token context window!
  • MiniMax M2.5 (minimax-m2.5): $0.15/$1.15, 204k ctx — 2025 flagship
  • MiniMax M2.7 (minimax-m2.7): $0.28/$1.20, 204k ctx — improved instruction following
• New provider: IBM (2 models — Granite 4 family):
  • IBM Granite 4.0 Micro (granite-4.0-micro): $0.02/$0.11, 131k ctx — ultra-cheap enterprise SLM
  • IBM Granite 4.1 8B (granite-4.1-8b): $0.05/$0.10, 131k ctx — strong reasoning + code
• New provider: ByteDance (3 models — Seed family):
  • Seed 1.6 Flash (seed-1.6-flash): $0.07/$0.30, 262k ctx — fast inference; multilingual
  • Seed-2.0-Mini (seed-2.0-mini): $0.10/$0.40, 262k ctx — balanced speed/quality
  • Seed-2.0-Lite (seed-2.0-lite): $0.25/$2.00, 262k ctx — higher quality tier
• Total: 196 → 209 models; 27 → 31 providers

• New provider: NVIDIA NIM (4 models):
  • Nemotron Super 49B (nemotron-super-49b): $0.10/$0.40, 131k ctx — Llama 3.3 Nemotron with extended reasoning
  • Nemotron 3 Nano 30B (nemotron-3-nano-30b): $0.05/$0.20, 262k ctx — MoE 3B active params, ultra-cheap
  • Nemotron 3 Super 120B (nemotron-3-super-120b): $0.09/$0.45, 262k ctx — MoE 12B active, strong
  • Nemotron Nano 9B (nemotron-nano-9b): $0.04/$0.16, 131k ctx — smallest Nemotron
• New provider: NousResearch (3 models):
  • Hermes 4 405B (hermes-4-405b): $1.00/$3.00, 131k ctx — flagship function-calling model
  • Hermes 4 70B (hermes-4-70b): $0.13/$0.40, 131k ctx — strong tool-use at mid-tier pricing
  • Hermes 3 Llama 3.1 405B (hermes-3-llama-3.1-405b): $1.00/$1.00, 131k ctx — flat-rate pricing
• Qwen 2.5 series (3 models):
  • Qwen2.5-72B (qwen-2.5-72b): $0.36/$0.40, 32k ctx — strong multilingual predecessor to Qwen3
  • Qwen2.5-7B (qwen-2.5-7b): $0.04/$0.10, 32k ctx — ultra-cheap compact model
  • Qwen2.5-Coder-32B (qwen-2.5-coder-32b): $0.66/$1.00, 32k ctx — SOTA open-weights code model
• DeepSeek V3.1 Terminus (deepseek-v3.1-terminus): $0.27/$0.95, 163k ctx — stronger reasoning than V3.2
• Total: 185 → 196 models; 25 → 27 providers

• New provider: Inflection AI
  • Inflection-3 Pi (inflection-3-pi): $2.50/$10.00 per Mtok, 8k ctx — conversational AI (Pi)
  • Inflection-3 Productivity (inflection-3-productivity): $2.50/$10.00, 8k ctx — enterprise tasks
• Qwen3 model expansion (4 new models):
  • Qwen3-8B (qwen3-8b): $0.05/$0.40 per Mtok, 40k ctx — smallest Qwen3 dense model
  • Qwen3-30B-A3B MoE (qwen3-30b-moe): $0.09/$0.45, 40k ctx — 3B active params, very cheap
  • Qwen3 Max (qwen3-max): $0.78/$3.90, 262k ctx — Alibaba's Qwen3 flagship with thinking mode
  • Qwen3 Coder (qwen3-coder): $0.22/$1.80, 262k ctx — coding specialist, 262k context
• Google Gemma 4 (2 new models):
  • Gemma 3 4B (gemma-3-4b): $0.04/$0.08, 131k ctx — ultra-compact Gemma 3
  • Gemma 4 27B Dense (gemma-4-27b): $0.12/$0.37, 262k ctx — Google's 2026 open-weights flagship
  • Gemma 4 27B MoE (gemma-4-27b-moe): $0.06/$0.33, 262k ctx — cheaper MoE variant
• Mistral additions (3 new models):
  • Mistral Saba (mistral-saba): $0.20/$0.60, 32k ctx — Arabic & multilingual specialist (24B)
  • Ministral 3B (ministral-3b): $0.10/$0.10, 131k ctx — smallest Mistral model, flat rate
  • Voxtral Small (voxtral-small): $0.10/$0.30, 32k ctx — Mistral's speech/voice processing model
• Total: 173 → 185 models; 24 → 25 providers

• New provider: Microsoft Azure AI Foundry
  • Phi-4 14B (phi-4): $0.065/$0.140 per Mtok, 16k ctx — SOTA SLM for reasoning/coding
  • Phi-4 Mini (phi-4-mini-instruct): $0.080/$0.350 per Mtok, 131k ctx — 3.8B compact model
  • Phi-4 Reasoning (phi-4-reasoning): $1.50/$6.00 per Mtok, 32k ctx — chain-of-thought model
• DeepSeek-R1-0528 (deepseek-r1-0528): $0.55/$2.19 per Mtok, 163k ctx — May 2026 R1 checkpoint; improved agentic + STEM reasoning; available on DeepSeek native API
• Qwen3-32B (qwen3-32b): $0.20/$0.60 per Mtok, 131k ctx — dense model, strong coding/math
• Qwen3-14B (qwen3-14b): $0.10/$0.30 per Mtok, 131k ctx — compact with excellent efficiency
• Total: 167 → 173 models; 23 → 24 providers

• DeepSeek V4 Flash (deepseek-v4-flash): $0.14/$0.28 per Mtok, 1M ctx — Cheapest DeepSeek model with massive 1M context window
• DeepSeek-R1 Distill Qwen-32B (deepseek-r1-distill-qwen-32b): $0.29/$0.29 per Mtok, 32k ctx — R1 reasoning distilled into Qwen2.5-32B; flat-rate pricing
• DeepSeek-R1 Distill Llama-70B (deepseek-r1-distill-llama-70b): $0.70/$0.80 per Mtok, 131k ctx — R1 reasoning on Llama-3.3-70B open-weights backbone
• Ministral 14B (ministral-14b): $0.20/$0.20 per Mtok, 262k ctx — Ministral 14B (Dec 2025); 262k ctx; step up from 8B at same flat-rate pricing
• Mistral Small 2603 (mistral-small-2603): $0.15/$0.60 per Mtok, 262k ctx — Mistral Small March 2026; 262k ctx; asymmetric pricing with cheap input
• Gemma 3 27B (gemma-3-27b): $0.08/$0.16 per Mtok, 131k ctx — Google's flagship open-weights Gemma 3 model; runs via Google AI Studio
• Gemma 3 12B (gemma-3-12b): $0.04/$0.13 per Mtok, 131k ctx — Compact Gemma 3; strong multilingual + coding capability
• New provider: Qwen / Alibaba Cloud (direct dashscope API)
  • Qwen Turbo (qwen-turbo): $0.03/$0.13 per Mtok, 131k ctx — fastest/cheapest
  • Qwen Plus (qwen-plus): $0.26/$0.78 per Mtok, 1M ctx — 1M context window
  • Qwen Max (qwen-max): $1.04/$4.16 per Mtok, 32k ctx — Alibaba flagship
  • Qwen3-235B (qwen3-235b): $0.45/$1.82 per Mtok, 131k ctx — 235B MoE open-weights
• Total: 157 → 167 models; 22 → 23 providers

• Mistral Pixtral Large (pixtral-large): $2.00/$6.00 per Mtok, 131k ctx — Mistral's multimodal vision flagship (Nov 2024); same price tier as Mistral Large 2
• Mistral Mixtral 8×22B (mixtral-8x22b): $2.00/$6.00 per Mtok, 65k ctx — Mistral's largest open-weights MoE (141B total params); 141B total params
• Mistral NeMo 12B (mistral-nemo): $0.02/$0.03 per Mtok, 131k ctx — Ultra-cheap Mistral+NVIDIA collaboration (Jul 2024); open-weights; among cheapest per-token
• Llama 3.2 11B Vision (llama-3.2-11b-vision): $0.18/$0.18 per Mtok, 128k ctx — Meta's multimodal vision model via Groq; compact image+text at flat rate
• Llama 3.2 90B Vision (llama-3.2-90b-vision): $0.90/$0.90 per Mtok, 128k ctx — Meta's large multimodal vision model via Groq; highest accuracy Llama 3.2 vision
• Total: 152 → 157 models

• Mistral Magistral Medium (magistral-medium): $2.00/$5.00 per Mtok, 40k ctx — Mistral's first reasoning flagship (Jun 2025); extended thinking, 40k output window
• Mistral Magistral Small (magistral-small): $0.50/$1.50 per Mtok, 40k ctx — Compact reasoning model; same price tier as Mistral Large 3 but with chain-of-thought
• xAI Grok-3 Mini Fast (grok-3-mini-fast): $0.60/$4.00 per Mtok, 131k ctx — High-speed reasoning tier; premium compute for latency-sensitive reasoning workloads
• Total: 149 → 152 models

• Claude Opus 4.6 (claude-opus-4-6): $5.00/$25.00 per Mtok, 128k ctx — Feb 2026 Opus 4 revision; expanded context at same price as 4.7
• Claude Opus 4.5 (claude-opus-4-5): $5.00/$25.00 per Mtok, 64k ctx — Nov 2025 mid-cycle Opus 4 revision
• Claude Opus 4.1 (claude-opus-4-1): $15.00/$75.00 per Mtok, 32k ctx — Original Opus 4 launch (Aug 2025); expensive legacy tier
• Claude Sonnet 4.5 (claude-sonnet-4-5): $3.00/$15.00 per Mtok, 64k ctx — Sep 2025 first Sonnet 4 release; 64k context window
• Cohere Command A (command-a): $2.50/$10.00 per Mtok, 256k ctx — Mar 2025 enterprise instruction-following model; 256k context; replaces Command R+
• Total: 144 → 149 models

• GPT-5 base (gpt-5): $1.25/$10.00 per Mtok, 128k ctx — Original GPT-5 release (Aug 2025); foundation of the GPT-5 family
• GPT-5.1 (gpt-5.1): $1.25/$10.00 per Mtok, 128k ctx — Nov 2025 GPT-5 update; improved quality at same price as base GPT-5
• GPT-5.2 (gpt-5.2): $1.75/$14.00 per Mtok, 128k ctx — Dec 2025 GPT-5 update; improved performance at moderate price premium
• GPT-5 Pro (gpt-5-pro): $15.00/$120.00 per Mtok, 272k ctx — Premium GPT-5 tier; ultra-high capability for most demanding tasks
• Grok-4 standard (grok-4): $3.00/$15.00 per Mtok, 256k ctx — xAI's Grok-4 (Jul 2025); predecessor to 4.3 and 4.20-reasoning variants
• Grok Code Fast (grok-code-fast): $0.20/$1.50 per Mtok, 256k ctx — xAI coding specialist; ultra-low cost for code generation and review tasks
• Total: 138 → 144 models

• DeepSeek V3.2 (deepseek-v3.2): $0.28/$0.40 per Mtok, 163k ctx — DeepSeek's latest V3 revision with reasoning support and flat-rate output pricing
• OpenAI open-source 120B on Groq (gpt-oss-120b-gq): $0.15/$0.60 per Mtok, 131k ctx — OpenAI's open-weight 120B model hosted on Groq's fast inference
• OpenAI open-source 20B on Groq (gpt-oss-20b-gq): $0.075/$0.30 per Mtok, 131k ctx — Cheaper 20B tier of the OpenAI open-source family on Groq
• Google Deep Research Pro (gemini-deep-research-pro): $2.00/$12.00 per Mtok, 65k ctx — Multi-step web research with grounding; Google's dedicated research model
• Total: 134 → 138 models

• OpenAI o3 Deep Research (o3-deep-research): $10.00/$40.00 per Mtok, 200k ctx — Multi-step web + document research model using o3 reasoning
• OpenAI o4-mini Deep Research (o4-mini-deep-research): $2.00/$8.00 per Mtok, 200k ctx — Affordable deep research for agentic research loops
• OpenAI GPT-5 Codex (gpt-5-codex): $1.25/$10.00 per Mtok, 272k ctx — GPT-5 coding-focused model with 272k context
• OpenAI Codex Mini Latest (codex-mini-latest): $1.50/$6.00 per Mtok, 200k ctx — Compact, cost-efficient coding model
• Total: 130 → 134 models

• Gemini 3 Flash Preview (gemini-3-flash-preview): $0.50/$3.00 per Mtok, 1M ctx — Google's latest speed+intelligence model with search/grounding
• xAI Grok 4.1 Fast (grok-4-1-fast): $0.20/$0.50 per Mtok, 2M ctx — New cheaper xAI tier; 6× less expensive than grok-4.3 for both reasoning and non-reasoning
• Total: 128 → 130 models

• Nebius AI (19th provider) — 5 models:
  • llama-3.1-8b-nb: $0.02/$0.06 per Mtok, 128k ctx — cheapest Llama 3.1 8B available
  • llama-3.3-70b-nb: $0.13/$0.40, 128k ctx
  • qwen3-235b-nb: $0.20/$0.60, 262k ctx — 262k context window
  • nemotron-ultra-253b-nb: $0.60/$1.80, 128k ctx — Nvidia 253B research model
  • deepseek-r1-0528-nb: $0.80/$2.40, 164k ctx
• Total: 108 → 113 models, 18 → 19 providers
• 2 new tests for Nebius provider

• providers command enhanced: now shows min input/output price and max context window per provider
• providers command adds --markdown flag (GitHub-flavored table) and --json flag with structured data

• Novita AI (18th provider) — 4 models:
  • llama-4-maverick-no: $0.27/$0.85 per Mtok, 1M context
  • llama-4-scout-no: $0.18/$0.59, 131k ctx
  • deepseek-r1-0528-no: $0.70/$2.50, 163k ctx
  • qwen3-235b-no: $0.20/$0.80, 40k ctx
• Mistral new models (3 additions):
  • devstral-small: $0.10/$0.30, 128k ctx — coding agent model
  • devstral: $0.40/$2.00, 256k ctx — full coding agent, 256k context
  • ministral-3-8b: $0.15/$0.15, 262k ctx — ultra-cheap, huge context, same in/out price
• Total: 101 → 108 models, 17 → 18 providers
• Homebrew tap updated to v0.1.16 (SHA256 verified)

• DeepInfra (16th provider) — 4 models:
  • llama-4-maverick-di: $0.15/$0.60 per Mtok, 1M token context — largest Llama 4 context available
  • llama-4-scout-di: $0.08/$0.30, 327k ctx
  • deepseek-r1-0528-di: $0.50/$2.15, 163k ctx
  • qwq-32b-di: $0.15/$0.40, 131k ctx (Qwen QwQ reasoning model)
• Lambda AI (17th provider) — 4 models:
  • llama-4-maverick-la: $0.05/$0.10 per Mtok — tied with gpt-5-nano as cheapest capable model!
  • llama-4-scout-la: $0.05/$0.10, 16k ctx
  • llama3.3-70b-la: $0.12/$0.30, 131k ctx
  • deepseek-r1-0528-la: $0.20/$0.60, 131k ctx (very cheap reasoning)
• Total: 93 → 101 models, 15 → 17 providers
• 3 new tests for DeepInfra and Lambda providers

• pyproject.toml now uses dynamic versioning from llm_prices.__version__ — fixes PyPI publish

• Tests no longer hardcode /workspace/projects/llm-prices as cwd — they now work correctly in CI (GitHub Actions) and any other environment.

• gpt-5-nano: $0.05/$0.40 per Mtok, 272k ctx — ultra-cheap GPT-5
• gpt-5-mini: $0.25/$2.00 per Mtok, 272k ctx — base GPT-5 mini tier
• o3-pro: $20.00/$80.00 per Mtok, 200k ctx — professional-grade reasoning
• o1-pro: $150.00/$600.00 per Mtok, 200k ctx — most expensive model in dataset
• Total: 89 → 93 models (same 15 providers)

• Blog post rewritten: new headline/story focused on GPT-5, Grok 4.3, o3 repricing. Updated all model counts (80 → 93), terminal examples, provider highlights, and real-world cost calculations to reflect the current model landscape.

• OpenAI GPT-5 series (4 models):
  • gpt-5.5: $5.00/$30.00 per Mtok, 1M context — current OpenAI flagship
  • gpt-5.4: $2.50/$15.00 per Mtok, 1M context
  • gpt-5.4-mini: $0.75/$4.50 per Mtok, 272k context
  • gpt-5.4-nano: $0.20/$1.25 per Mtok, 272k context
• xAI Grok 4 series (2 models):
  • grok-4.3: $1.25/$2.50 per Mtok, 1M context — current Grok flagship
  • grok-4.20-reasoning: $1.25/$2.50 per Mtok, 2M context
• Total: 83 → 89 models (same 15 providers)
• Tests: 54 → 57 (test_gpt5_models_present, test_o3_repriced, test_grok4_present)

• o3: corrected from $10/$40 to $2/$8 per Mtok (OpenAI repriced ~5× cheaper)

• Google Gemini 3.1 Pro Preview ($2.00/$12.00, 1M ctx) — multimodal+agentic flagship preview
• Google Gemini 3.1 Flash-Lite ($0.25/$1.50, 1M ctx) — cost-efficient Gemini 3 model
• Google Gemini 2.5 Flash-Lite ($0.10/$0.40, 1M ctx) — smallest/cheapest Gemini 2.5
• Total: 80 → 83 models (same 15 providers)
• Tests: 51 → 54 (claude4_pricing_correct, gemini3_models_present, gemini_flash_pricing_updated)

• Claude Opus 4.7: corrected from $15/$75 to $5/$25 per Mtok; context 200k → 1M tokens (Anthropic official docs: claude-opus-4-7 is the affordable flagship, not legacy Opus 3 pricing)
• Claude Sonnet 4.6: context window 200k → 1M tokens
• Claude Haiku 4.5: corrected from $0.80/$4.00 to $1.00/$5.00 per Mtok
• Gemini 2.5 Flash: corrected from $0.15/$0.60 to $0.30/$2.50 per Mtok (hybrid reasoning model with thinking budgets — output price reflects that)
• Gemini 2.5 Pro: updated notes (tiered pricing: $1.25/$10 ≤200k, $2.50/$15 >200k)

• AI21 Labs (15th provider): Jamba Mini 1.7 ($0.20/$0.40, 256k ctx) and Jamba Large 1.7 ($2.00/$8.00, 256k ctx). Hybrid SSM+Transformer architecture.
• Total: 78 → 80 models, 14 → 15 providers
• Tests: 50 → 51 (test_ai21_models_present)

• test_list_csv: use csv.reader instead of naive split(",") to handle quoted fields that contain commas in the notes column

• Groq: Kimi K2 (kimi-k2-gq, $1.00/$3.00, 262k ctx) and Qwen3 32B (qwen3-32b-gq, $0.29/$0.59)
• Mistral Large 3 (mistral-large-3): $0.50/$1.50 per Mtok, 262k context — 4× cheaper than Large 2 with 2× the context window. Major price cut for the flagship model.
• Mistral Medium 3 (mistral-medium-3): $0.40/$2.00 per Mtok, 131k context (new family)
• Mistral Small 3.2 (mistral-small-3-2): $0.06/$0.18 per Mtok, 131k context (cheaper and larger context than the older Small 3)
• Total: 73 → 78 models, same 14 providers
• Tests: 48 → 50 (test_mistral_large3_present, test_groq_new_models)

• Amazon Bedrock provider: 5 models (nova-micro-br, nova-lite-br, nova-pro-br, nova-premier-br, nova-2-lite-br) Amazon's own Nova foundation models via Bedrock on-demand pricing (US East). Nova Micro has a 128k context at just $0.035/Mtok input — one of the cheapest options. Nova Premier offers a 1M token context window.
• Total: 68 → 73 models across 14 providers
• Test suite: 47 → 48 tests (test_bedrock_models_present)

• SambaNova provider: 5 models (llama-4-maverick-sb, llama-3.3-70b-sb, deepseek-v3-sb, minimax-m2-5-sb, gemma-3-12b-sb) Ultra-fast inference on SambaNova RDU (Reconfigurable Dataflow Unit) silicon. Includes Llama 4 Maverick — the latest Meta flagship (17B MoE). Prices sourced from the live SambaNova API (api.sambanova.ai/v1/models).
• Total: 63 → 68 models across 13 providers
• Test suite: 46 → 47 tests (test_sambanova_models_present)

• Cerebras provider: 3 models (llama-3.3-70b-cb, llama-3.1-8b-cb, qwen3-32b-cb) Ultra-fast inference on custom Cerebras silicon (wafer-scale chips)
• Total: 60 → 63 models across 12 providers

• top command: show N cheapest models for a given token workload (llm-prices top 5 --in 5000 --out 1000 --markdown)
• Tests expanded from 39 → 45 (6 new tests for top command)

• Test suite expanded from 26 → 39 tests
• Added coverage for: --markdown output, --csv output, compare --markdown, budget command, new providers (Together/Fireworks/Perplexity), model count assertions, sort order verification

• Perplexity AI provider: 4 models (sonar, sonar-pro, sonar-reasoning-pro, sonar-deep-research) with notes about per-request search fees
• Total: 56 → 60 models across 11 providers

• Fireworks AI provider: 6 models (deepseek-v4-pro-fw, deepseek-v3-fw, kimi-k2-fw, llama-3.1-70b-fw, llama-3.1-8b-fw, mixtral-8x7b-fw)
• DeepSeek V4 Pro on Together AI (deepseek-v4-pro-together)
• Total: 49 → 56 models across 10 providers

• Together AI provider: 6 models (qwen3-235b, kimi-k2, llama-3.3-70b-turbo, qwen3.5-9b, deepseek-r1-together, deepseek-v3-together)
• list --markdown: output as GitHub-flavored Markdown table (paste into READMEs/docs)
• list --csv: output as CSV (import into spreadsheets, databases)
• compare --markdown: output comparison as Markdown table
• Total: 43 → 49 models across 9 providers

• OpenAI: gpt-4.1, gpt-4.1-mini, gpt-4.1-nano (up to 1M context window)
• Google: gemini-2.5-flash
• Anthropic: claude-3-7-sonnet; versioned model IDs (claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5)
• Groq: llama-4-scout, llama-4-maverick
• Total: 36 → 43 models

• Corrected project URLs in pyproject.toml to point to actual GitHub repo

• Initial release
• list command: show all models with pricing, filterable by provider/name, sortable
• calc command: calculate exact cost for a model + token count
• compare command: side-by-side cost comparison of multiple models
• budget command: how many API calls fit within a dollar budget
• providers command: list available providers
• JSON output mode (--json) for all commands
• Fuzzy model name matching (normalises hyphens, underscores, case)
• 36 models across 8 providers: OpenAI, Anthropic, Google, Mistral, Groq, Cohere, DeepSeek, xAI
• Zero runtime dependencies (Python stdlib only)
• Donation tip jar in --help output and README
• 26-test suite (100% passing)
• GitHub Actions CI + PyPI Trusted Publishing workflow