Best Nous DeepHermes 3 8B Alternatives & Competitors in 2026

OpenAI Sep 29, 2026 flagship reasoning & coding model (superseding GPT-6 Sol) with $2/$10 base pricing and $0.10/1M (95% off) official cached input.

Audit: ⚠️ 272K Context Cliff: Standard $2.00/$10.00 (Cache $0.10) applies up to ≤272K tokens. Prompts exceeding >272K tokens jump to $4.00 input / $15.00 output per 1M tokens.

OpenAI Sep 3, 2026 ultra-frontier deep reasoning flagship ($10/$50 per 1M) with $1.00/1M (90% off) official cached input.

Audit: Internal reasoning tokens are billed at the $50/1M output rate. Always pin system prompts to utilize the $1.00/1M (90% off) cached input rate.

OpenAI Sep 22, 2026 high-throughput compact model priced 100x cheaper than GPT-6 Astra ($0.10/$0.50, Official Cache Read $0.01/1M).

Audit: <=272K: $0.10/$0.50 ($0.01 cache, $0.125 write); >272K full-request cliff: $0.20/$0.75 ($0.02 cache).

Anthropic Sep 22, 2026 flagship coding & autonomous agent model ($4/$20 per 1M, 20% cheaper than Opus 5 ($5/$25 -> $4/$20)) with $0.20/1M (95% off) cache read.

Audit: Adaptive thinking billed at $20/1M output rate ($8/$40 in 2x Fast mode; US-only +10%).

Anthropic Sep 1, 2026 heavy research & synthesis architecture ($10/$50 per 1M) featuring an industry-leading 97.5% cache read discount ($0.25/1M).

Audit: Un-cached multi-turn research loops burn $10/$50 rapidly; enable Prompt Caching immediately to lock in the $0.25/1M (97.5% off) read rate.

Anthropic Sep 28, 2026 production coding workhorse ($2/$10 per 1M). Official cache read cut 50% from $0.20 to $0.10/1M effective Oct 7, 2026.

Audit: Extended thinking tokens bill at Sonnet 5.5 standard $10/1M output rate; cache read cut 50% to $0.10/1M on Oct 7.

Anthropic Oct 7, 2026 ultra-fast model: $0.10/$0.50 (Cache $0.01/1M) for prompts ≤100K tokens; jumps 5x to $0.50/$2.50 (Cache $0.05) for >100K tokens.

Audit: <=100K: $0.10/$0.50 ($0.01 cache); >100K (including cached tokens) jumps 5x to $0.50/$2.50 ($0.05 cache) for the full request.

Google Oct 1, 2026 multimodal flagship currently at $2/$10 (Cache $0.10/1M) Time-Limited Intro Price through Nov 30, 2026; increases to $4/$20 (Cache $0.20) on Dec 1, 2026.

Audit: ⚠️ Time-Limited Intro Pricing Alert: The $2.00/$10.00 (Cache $0.10) rate expires on 2026-11-30! Starting 2026-12-01, standard pricing doubles to $4.00/$20.00 (Cache $0.20).

Google Gemini 3.8 Flash high-speed multimodal model ($0.75/$3.75 intro thru 2026-12-31, Official Cache Read $0.075/1M [90% off], reprices to $1.50/$7.50 in 2027).

Audit: ⚠️ Intro Rate Expiration: Official $0.75/$3.75 (Cache $0.075/1M) introductory pricing ends 2026-12-31, increasing to $1.50/$7.50 in Jan 2027.

DeepSeek-V4 Pro (0813) 1.6T MoE flagship: Off-Peak $0.66/$1.98 ($0.022/1M KV cache hit, 96.7% off); Weekday UTC Peak (01-04 & 06-10) 2x at $1.32/$3.96 ($0.044 cache).

Audit: Off-Peak: $0.66/$1.98 ($0.022 cache); Weekday UTC Peak (01-04 & 06-10): $1.32/$3.96 ($0.044 cache). Chinese statutory holidays count as all-day Off-Peak.

DeepSeek-V4.1-Flash-1017 high-speed coding & triage model: Off-Peak $0.15/$0.60 ($0.003/1M KV cache hit, 98% off); Peak $0.30/$1.20 ($0.006 cache).

Audit: Off-Peak: $0.15/$0.60 ($0.003 cache); Weekday UTC Peak: $0.30/$1.20 ($0.006 cache). Search/Fetch API ($0.75/1K) begins billing 2026-10-31T16:00:00Z (Nov 1 00:00 Beijing).

OpenAI GPT-5.6 Terra router catalog tier ($2.00/$12.00 per 1M, 400K flat context window, $0.20 cache read).

Audit: Flat context pricing ($2.00/$12.00, Cache $0.20/1M); consider migrating to GPT-6.1 Sol ($2.00/$10.00, Cache $0.10/1M) for 95% cache savings.

OpenAI Daybreak Program specialized defensive cybersecurity & vulnerability auditing model ($12.50/$75.00 per 1M, restricted authorization).

Audit: Requires OpenAI Daybreak Program enterprise security clearance. High $12.50/$75.00 token rate; enable $1.25/1M prompt caching for codebase scans.

Flagship GPT-5 series general model with configurable reasoning_effort from none to xhigh.

Audit: Official gpt-5.5-chat-latest cached input rate is $0.50/1M (90% off $5.00 input). Migrate to GPT-6.1 Sol ($2/$10, Cache $0.10) to cut spend by 50%.

OpenAI GPT-5.4 nano high-throughput classification & routing model ($0.20/$1.25 per 1M, Official Cache Read $0.02/1M [90% Off], 128K context window).

Audit: Enable Prompt Context Caching & Batch API to reduce repeated system prompt input token burn by 50%-97.5%.

OpenAI specialized code generation & repository refactoring model ($1.75/$14.00 per 1M, Official Cache Read $0.175/1M).

Audit: High $14.00/1M output rate; pin repository context in prefix cache ($0.175/1M read, 90% off) or migrate to GPT-6.1 Sol ($2/$10, Cache $0.10).

xAI frontier reasoning & coding model (500K context): <200K at $2/$6 ($0.50 cache); >=200K at $4/$12 ($1.00 cache). No batch discount.

Audit: <200K: $2.00/$6.00 ($0.50 cache); >=200K (including cached tokens) full-request 2x cliff: $4.00/$12.00 ($1.00 cache). Not eligible for Batch discount (0%); Priority 2x; US endpoint +10%.

xAI Grok 4.3 high-speed reasoning model: $1.25/$2.50 (Cache $0.20/1M) short context; jumps to $2.50/$5.00 (Cache $0.40/1M) on long context.

Audit: <200K: $1.25/$2.50 ($0.20 cache); >=200K full-request 2x cliff: $2.50/$5.00 ($0.40 cache). Batch discount is 20%; Priority tier 2x; US endpoint +10%.

Meta Llama 4 Scout 109B open-weight MoE model ($0.11/$0.34 per 1M, $0.022/1M cache read [80% off], 10M token context window).

Audit: Enable Prompt Context Caching & Batch API to reduce repeated system prompt input token burn by 50%-97.5%.

DeepSeek-R1 (deepseek-reasoner) open-weight reinforcement-learning reasoning model ($0.55/$2.19 per 1M, $0.14/1M disk KV cache hit, 128K context window).

Audit: Enable Prompt Context Caching & Batch API to reduce repeated system prompt input token burn by 50%-97.5%.

Flint is an open-weight mixture-of-experts model from Recursive, with about 2.5B active parameters out of 40B total, built by Recursive's autoresearch system Eureka Machine for cod

Audit: Auto-discovered via live global telemetry registry.

MiniMax-M3.1-Flash is a multimodal model from MiniMax. Compared with M3, it delivers significant improvements across multiple benchmarks, particularly in complex agentic tasks, cod

Audit: Auto-discovered via live global telemetry registry.

DeepSeek Oct 17, 2026 Pareto coding update ("Wish-Driven Programming"): 77.0% on DeepSWE v1.1, 60.5% Automation-Bench, $0.15/$0.60 Off-Peak ($0.003 Cache Hit) + Search/Fetch API ($0.75/1K).

Audit: Official off-peak token rate is $0.15/$0.60 with $0.003/1M disk-KV cache hit. Built-in Search & Fetch API calls are billed separately at $0.75 per 1,000 requests.

Glimpse is a multimodal visual reasoning model from Elorian, built for visual understanding and complex visual reasoning in the physical world. It reasons directly over images and

Audit: Auto-discovered via live global telemetry registry.

Xing4.0-29B-A4B is an open-weight mixture-of-experts language model from China Telecom's AI team (TeleAI) and the successor to the TeleChat series. It has 29B total parameters with

Audit: Auto-discovered via live global telemetry registry.

Beam 1.0 is Reflection's open-weight reasoning model for coding and long-horizon agentic workflows. It uses a sparse mixture-of-experts architecture with 501B total and 23B active

Audit: Auto-discovered via live global telemetry registry.

Mercury 3 is a diffusion-based language model from Inception, and the lab's most capable model to date. It is suited for agentic workflows and real-time applications where latency

Audit: Auto-discovered via live global telemetry registry.

Moondream 3 Preview is a vision-language model from Moondream with 9B total parameters and 2B active (mixture-of-experts). It handles visual question answering, captioning, OCR, an

Audit: Auto-discovered via live global telemetry registry.

Moondream 3.1 is a vision-language model from Moondream, the successor to Moondream 3 Preview, with 9B total parameters and 2B active (mixture-of-experts). It handles visual questi

Audit: Auto-discovered via live global telemetry registry.

Muse Spark Open is a coding-optimized open-weight model. Trained specifically for agentic coding workflows, it delivers higher first-attempt accuracy, reliable tool calling, and de

Audit: Auto-discovered via live global telemetry registry.

NVIDIA Nemotron 3.5 Super VL is an open multimodal mixture-of-experts model from NVIDIA, with 12B active parameters out of 120B total. It takes text, images, documents, GUI screens

Audit: Auto-discovered via live global telemetry registry.

NVIDIA Nemotron 3.5 Super VL is an open multimodal mixture-of-experts model from NVIDIA, with 12B active parameters out of 120B total. It takes text, images, documents, GUI screens

Audit: Auto-discovered via live global telemetry registry.

Pareto is a multimodal composite model built for research, coding, and agentic workflows, while delivering frontier-level performance across a broad range of general-purpose tasks.

Audit: Auto-discovered via live global telemetry registry.

This model always redirects to the latest model in the Pareto family.

Audit: Auto-discovered via live global telemetry registry.

Pareto is a multimodal composite model built for research, coding, and agentic workflows, while delivering frontier-level performance across a broad range of general-purpose tasks.

Audit: Auto-discovered via live global telemetry registry.

LFM2.5-24B-A2B is a mixture-of-experts reasoning model from Liquid AI with 24B total and 2.45B active parameters. It always reasons before answering and is built for agents that ca

Audit: Auto-discovered via live global telemetry registry.

Summarize is an abstractive summarization model from ScaleDown. It condenses long text and documents into a fluent, faithful rewrite rather than lifting sentences from the source,

Audit: Auto-discovered via live global telemetry registry.

Extract is a description-driven entity extraction model from ScaleDown. Instead of a fixed set of entity types, callers describe the fields they want in plain English through `prov

Audit: Auto-discovered via live global telemetry registry.

Muse Wonder is a frontier reasoning model from Meta, designed for complex personal and professional agentic tasks. It accepts text, images, video, audio, and PDF documents and retu

Audit: Auto-discovered via live global telemetry registry.

Step 5 Preview is StepFun's flagship model for agentic work, built on a sparse Mixture-of-Experts architecture (27B active / 600B total parameters). It performs strongly in softwar

Audit: Auto-discovered via live global telemetry registry.

Nano Banana 2.1 (Gemini Nano Banana 2.1) is Google's image generation and editing model on the Flash tier, succeeding Nano Banana 2 and Nano Banana Pro. It improves product reconte

Audit: Auto-discovered via live global telemetry registry.

Mistral Large 4 is a frontier multimodal (text and image input) model from Mistral AI built for reasoning, coding, and agentic workloads. It offers a 1M-token context window with u

Audit: Auto-discovered via live global telemetry registry.

Ling 3.1 Flash is a hybrid reasoning mixture-of-experts model from inclusionAI, with 25B active parameters out of 560B total.

Audit: Auto-discovered via live global telemetry registry.

GPT-6.1 Sol Pro is the same underlying model as [GPT-6.1 Sol](https://openrouter.ai/openai/gpt-6.1-sol), served with `reasoning.mode` set to `pro` for higher-quality responses on c

Audit: Auto-discovered via live global telemetry registry.

Perceptron Mk1.5 is Perceptron's embodied reasoning model for physical agents. It accepts text, image, video, and audio input, and answers with text plus optional structured annota

Audit: Auto-discovered via live global telemetry registry.

Ember-1 is a specialized reasoning model from Fireworks Research, built on [Kimi K3](https://openrouter.ai/moonshotai/kimi-k3). It is designed to make every token go further: it pr

Audit: Auto-discovered via live global telemetry registry.

GLM-5.3-Prime is the high-speed variant of Z.ai's GLM-5.3, inheriting its full capabilities while delivering 1.5–2× the output throughput through inference acceleration. It support

Audit: Auto-discovered via live global telemetry registry.

Qwen3.8 Max Prime is a higher-throughput variant of Qwen3.8 Max from Alibaba's Qwen team, served as a separate SKU at a higher price point. It accepts text, image, and video...

Audit: Auto-discovered via live global telemetry registry.

Aion 3.5 Mini is a multi-model roleplaying and storytelling system from AionLabs, built on the GLM family of models. It is the smaller, lower-cost sibling of Aion 3.5 and uses...

Audit: Auto-discovered via live global telemetry registry.

Aion 3.5 is a multi-model roleplaying and storytelling system from AionLabs, built on the GLM family of models. It uses a collaborative generation process in which multiple special

Audit: Auto-discovered via live global telemetry registry.

Solar Mini 4 is Upstage's compact, cost-efficient language model, a 35B-parameter mixture-of-experts with 3B active parameters and a 524K context window. It is built for agentic us

Audit: Auto-discovered via live global telemetry registry.

Command A+ is Cohere's flagship model for enterprise agentic workflows. It accepts text and image inputs with a 192K context window, supports native tool calling with strict tool s

Audit: Auto-discovered via live global telemetry registry.