Claude alternatives worth considering
The credible alternatives to Claude in August 2026 - ChatGPT, Gemini, Grok, DeepSeek, Copilot and the open-weight models - with prices and trade-offs.
The best alternative to Claude depends on which of Claude's weaknesses is blocking you. If it is cost, DeepSeek V4-Pro and Google's Gemini 3.7 Flash are an order of magnitude cheaper. If it is generative media, only OpenAI and Google are candidates, because Claude generates no images, video, audio or music at all. If it is self-hosting, Meta's Muse Glimmer 30B under Apache 2.0 is the cleanest genuinely-open option at a useful size.
This page surveys the credible field as of August 2026, with prices attributed to each vendor's own page and licence claims stated only where a licence has actually been published.
#The short answer
- Cheapest capable
- DeepSeek V4-Pro - $0.435 / $0.87 per MTok
- Best for generative media
- Google (Veo 3.1, Imagen, Lyria 3) or OpenAI (Sora 2, GPT-Image-2)
- Best for very long documents
- Gemini 3.7 Flash - GDM-MRCR v2 97.0%
- Best open-weight, self-hostable
- Meta Muse Glimmer 30B (Apache 2.0)
- Best inside Microsoft 365
- Microsoft 365 Copilot, from about $18/user/mo
- Best search-grounded answers
- Perplexity Pro, $20/mo
#The field at a glance
Nine credible alternatives, with the single number that matters most for each. API prices are per million tokens; consumer prices are per month.
| Alternative | Frontier model | API in / out | Consumer entry | Pick it for |
|---|---|---|---|---|
| Anthropic (baseline) | Fable 5 / Opus 5 / Sonnet 5 | $10/$50 · $5/$25 · $2/$10 | Free, then $20 Pro | Agentic coding, novel reasoning, prose |
| OpenAI | GPT-5.6 Sol / Terra / Luna | Luna $0.20/$1.20; Terra $2/$12 | Free (unlimited text), Go $8 | Media generation, computer use, lowest unit cost |
| Gemini 3.7 Flash | $0.75/$3.75 (introductory) | Free, AI Plus ~$5–8 | Long-context retrieval, speed, video and music | |
| xAI | Grok 4.6 (500k context) | $2.00/$6.00 | Free, SuperGrok $30 | Frontier-class intelligence at a third of Sol's price |
| DeepSeek | V4-Pro (1M context) | $0.435/$0.87 | API-first | The sharpest price pressure in the market |
| Mistral | Mistral Medium 3.5, Large 3 | $1.50/$7.50 · $0.50/$1.50 | Le Chat Pro €/$14.99 | EU data residency and private deployment |
| Microsoft Copilot | OpenAI models, resold | Seat-based, not per-token | ~$18/user/mo business | Being already inside Word, Excel and Teams |
| Perplexity | Orchestrates others | Credit-based | Free, Pro $20 | Cited, search-grounded answers |
| Open weights | Muse Glimmer 30B, Kimi K3, GLM-5.3 | Free to self-host, or cheap hosted | - | Offline, air-gapped or cost-floor deployments |
#The two direct replacements: ChatGPT and Gemini
ChatGPT is the alternative most Claude users actually consider. OpenAI's GPT-5.6 family - Sol, Terra and Luna, released 9 July 2026 - all carry a 1.05M-token context window. It wins ARC-AGI-2 (92.5% to Opus 5's 90.4%, both verified independently by ARC Prize), OSWorld 2.0 for computer use (62.6% to 54.8%), Agents' Last Exam (52.7% to Fable 5's 40.5%) and holds the highest published Terminal-Bench 2.1 score at 91.9% in its Ultra multi-agent mode. It also generates images and video, which Claude cannot, and ChatGPT Free offers unlimited text chat where the Claude free plan is metered. The full breakdown, including where Claude wins, is on Claude vs ChatGPT.
One caution for developers: OpenAI's own marketing pricing page and its developer documentation publish different output rates for GPT-5.6 Sol and cannot be reconciled as a context-tier difference. Price Sol against the developer docs card rather than the marketing page, and confirm before you commit volume.
Gemini is the better alternative if your problem is documents, latency or budget. Gemini 3.7 Flash - Google's flagship since 13 August 2026, and notably a Flash model, with the Pro line still in preview - reports 97.0% on GDM-MRCR v2 long-context retrieval against Claude Sonnet 5's 81.5%, runs at roughly 3,901 output tokens per second against Fable 5's 75, and costs $0.75 in / $3.75 out per million tokens on Google's published rate card. Those rates are introductory through 31 December 2026 and double on 1 January 2027. Detail on Claude vs Gemini.
#The value-frontier options: Grok and DeepSeek
Grok 4.6 (xAI, shipped 12 August 2026) is the most underrated name on this list. It carries a 500,000-token context window and a February 2026 knowledge cutoff at $2.00 in / $6.00 out per million tokens per docs.x.ai, and it sits joint-fifth on the independent Artificial Analysis Intelligence Index at 61 - matching GPT-5.6 Sol at roughly a third of the cost. xAI also sells a voice API and an image/video API, so unlike Claude it is not media-blind. Consumer plans run Free, SuperGrok at $30/month and SuperGrok Heavy at $300/month. Note that several outlets and Artificial Analysis now label the company SpaceXAI following a reported combination with SpaceX; the corporate detail is not independently confirmed here.
DeepSeek V4-Pro is the price floor at the frontier: $0.435 in / $0.87 out per million tokens with a 1M-token context, per deepseek.ai/pricing, alongside V4-Flash at $0.14/$0.28. Against Fable 5 at $10/$50 that is roughly 23× cheaper on input and 57× cheaper on output. Announced peak-hour surcharges are reportedly not currently active. If your workload is high-volume classification, extraction or summarisation rather than long agentic runs, the honest advice is that Claude's premium is very hard to justify - compare it against Haiku 4.5 at $1/$5, which is Anthropic's cheapest answer and still more than double DeepSeek's rate.
#The sovereignty and deployment option: Mistral
Mistral's differentiator is not frontier capability - it is where the model runs and who controls it. The API spans Mistral Medium 3.5 ($1.50/$7.50), Mistral Large 3 ($0.50/$1.50), Mistral Small 4 ($0.15/$0.60), the Ministral 3 family at 3B/8B/14B ($0.10–$0.20 flat), Codestral for code ($0.30/$0.90), Voxtral for audio and OCR 4.1 at $4 per 1,000 pages, per mistral.ai/pricing/api. Le Chat plans run Free with $10/month of API credits, Pro at $14.99/month ($5.99 for students), Team at $24.99/user/month, and Enterprise with private deployments.
Choose Mistral when the requirement is EU data residency, on-premise or air-gapped hosting, or a procurement process that will not accept a US-only inference footprint. Anthropic's answer to part of this is Enterprise-only controls - custom data retention, customer-managed encryption keys, US-only inference and IP allowlisting - but it does not offer on-prem deployment at all.
#The distribution plays: Microsoft Copilot and Perplexity
Microsoft 365 Copilot is not a lab; it is a channel that puts OpenAI models where the work already happens. Business pricing on annual billing runs about $18/user/month promotionally through 30 September 2026 (regular $21), $23.50 for Business Standard with Copilot and $32 for Business Premium with Copilot, per Microsoft's pricing page; monthly billing is materially higher. Its advantage is access to Office and Graph data, not model quality. This is the distribution gap Anthropic has no equivalent of - Claude ships Microsoft 365 add-ins, but as something you install rather than something already in the ribbon.
Perplexity competes on search grounding and citations rather than on owning a model - it orchestrates ChatGPT, Gemini, Claude and NVIDIA Nemotron and lets paid users pick per query. Free, Pro at $20/month with deep research and a Computer agent, Max at $200/month, per perplexity.ai/hub/pricing. If what you actually want is answers with sources you can click, this beats any raw model, including Claude with web search enabled.
#Open-weight alternatives - and one you should not call open
The open and cheap tier is now within roughly three Intelligence Index points of the closed frontier, which is the single most important change of 2026.
| Model | Licence | Standing | Best for |
|---|---|---|---|
| Meta Muse Glimmer 30B | Apache 2.0, on Hugging Face since 10 Aug 2026 | Distilled from Muse Spark; targets single-GPU local agentic use | Genuinely open, self-hosted, air-gapped work |
| Meta Muse Spark 1.2 | Open weights announced, not yet shipped | arena.ai Elo 1498 - above Claude Opus 5 at 1493; Intelligence Index 57 at $0.40 blended | Prose quality at a fraction of frontier cost |
| Kimi K3 (Moonshot) | Open-weight lineage | Intelligence Index 60; arena.ai Elo 1490 | Frontier-adjacent reasoning on a budget |
| GLM-5.3 (Z AI / Zhipu) | Open-weight lineage | Intelligence Index 60; arena.ai Elo 1479 | Cheap hosted inference; Mistral resells the 5.2 generation |
| Qwen3.8-Max (Alibaba) | Licence not disclosed | 2.4T-parameter MoE, 95B active, 1M context, $2/$6; Index 58 | Capable and cheap - but do not treat it as open source |
Alibaba announced weights for Qwen3.8-Max and Qwen3.8-27B on Hugging Face and ModelScope but has not published the licence terms as of 21 August 2026. Until it does, "open weights" is a claim about availability, not about what you are permitted to do. Do not build a compliance story on it, and treat any article calling it open source as unverified.
The headline embarrassment for Anthropic here is Muse Spark 1.2: on arena.ai's human-preference board it ranks fourth at 1498 Elo, ahead of Claude Opus 5 at 1493, ahead of every Gemini entry and ahead of every GPT entry. Claude still holds the top slot with Fable 5 at 1507 and six of the top twelve places, so the family lead is intact - but the newest Opus is not the best-liked model even within Anthropic's own line, sitting below Opus 4.6 and Opus 4.7.
Anthropic ships no open weights of any kind. For air-gapped, sovereign or cost-floor deployments, Claude is simply not in the running, and no amount of API discounting changes that.
#Which alternative should you choose?
By need, not by brand.
| Your need | Pick | Why, specifically |
|---|---|---|
| Cheapest capable model | DeepSeek V4-Pro | $0.435/$0.87 with 1M context. GPT-5.6 Luna at $0.20/$1.20 is cheaper on input but has a smaller quality-per-dollar story on hard tasks |
| Generative media | Google, or OpenAI | Veo 3.1, Imagen and Lyria 3 versus Sora 2 and GPT-Image-2. Claude has no offering at any price |
| Very long documents | Gemini 3.7 Flash | GDM-MRCR v2 97.0% against Sonnet 5's 81.5%. Window size and retrieval accuracy are different things |
| Offline or self-hosted | Meta Muse Glimmer 30B | Apache 2.0, 30B, built for single-GPU local agentic use. The only clean licence at a useful size |
| EU residency or on-prem | Mistral | Private deployments and an EU-based footprint; Anthropic offers no on-prem option |
| Already inside Microsoft 365 | Microsoft 365 Copilot | About $18/user/month reaches Office and Graph data no external assistant can see |
| Already inside Google Workspace | Gemini | Default placement in Search, Gmail, Docs and Android beats a seven-point index gap |
| Search-grounded answers with citations | Perplexity Pro | $20/month, model choice per query, built around sourcing rather than fluency |
| Frontier intelligence at lower cost | Grok 4.6 | Intelligence Index 61 at $2/$6 - matches GPT-5.6 Sol for roughly a third the price |
| Long agentic coding runs on a real repo | Stay with Claude | Top of the official Terminal-Bench 2.1 board at 83.8%; the 80.3% (Mythos 5, invitation-only) SWE-bench Pro figure is not from a model you can buy. See Claude for developers |
Two closing notes. First, these are not exclusive choices - API keys are cheap and most serious teams route different task types to different models. Second, before switching for price alone, check what you would actually save: Anthropic's Batch API applies a flat 50% discount and prompt-cache reads cost 0.1× base input, both explained on the Claude API page, and Claude Pro at $200/year works out cheaper than ChatGPT Plus, which has no annual discount. The full plan mechanics are on the pricing page, and our verdict on whether Claude earns its premium is on the Claude review.
#Frequently asked questions
What is the cheapest good alternative to Claude?
DeepSeek V4-Pro at $0.435 in and $0.87 out per million tokens, with a one-million-token context window. That is roughly 23 times cheaper on input than Claude Fable 5. GPT-5.6 Luna at $0.20/$1.20 is cheaper still on input, and Gemini 3.7 Flash sits between them at $0.75/$3.75.
Is there an open-source alternative to Claude?
Meta's Muse Glimmer 30B is released under Apache 2.0 and is the cleanest genuinely-open option at a size you can run on one GPU. Kimi K3 and GLM-5.3 also have open-weight lineage. Anthropic itself publishes no open weights of any kind.
Is Qwen3.8-Max open source?
Unconfirmed. Alibaba announced weights on Hugging Face and ModelScope but has not disclosed the licence terms as of 21 August 2026. Until the licence is published, treat availability and permission as separate questions and do not build a compliance argument on it.
Which alternative is best for images and video?
Google or OpenAI. Google ships Veo 3.1 for video, Imagen for images and Lyria 3 for music; OpenAI ships Sora 2 and GPT-Image-2. Claude generates none of these and cannot accept video or audio as input, so it is not a candidate for media work.
Does any model beat Claude on human preference?
One does. Meta's Muse Spark 1.2 ranks fourth on arena.ai at 1498 Elo, above Claude Opus 5 at 1493. Claude Fable 5 still holds first place at 1507, and Anthropic occupies six of the top twelve positions on that board.
Should I switch away from Claude to save money?
Only if your workload is high-volume and repetitive. For classification, extraction and summarisation the price gap is impossible to justify. For long agentic coding runs where a failed task costs more than the tokens, Claude Code's lead at the top of the official Terminal-Bench 2.1 board still pays for itself.
Claude prices verified 21 August 2026 against claude.com/pricing. Competitor prices come from each vendor's own page, linked in the sections above. Independent rankings are from Artificial Analysis, arena.ai and ARC Prize. This market re-prices monthly - treat every figure here as a starting point for your own check, not a quote.