57 AI and ML Tools Pricing Compared (March 2026)

LLM API prices dropped 50–75% between 2024 and 2026. The tools you budgeted for last year cost materially less today — if you're still working from old numbers, you're either overspending or over-provisioning.

We compiled current pricing for 57 AI and ML tools across 10 categories, all verified March 2026. This covers LLM APIs, AI coding assistants, image generation, vector databases, ML platforms, AI agents, speech/audio, AI search/RAG, computer vision, and AI observability.

---

LLM APIs — Current Pricing

The token price wars continued through 2025 and into 2026. Here's where pricing stands:

Model	Input (per 1M tokens)	Output (per 1M tokens)	Notes
Mistral Small 3.1	$0.03	$0.11	Cheapest production-grade option; open weights
GPT-4o Mini	$0.15	$0.60	Best for high-volume, cost-sensitive tasks
Google Gemini 2.5 Flash	$0.30	$2.50	Cheapest Google flagship; 50% batch discount
Claude Haiku 4.5	$1.00	$5.00	Anthropic's fastest, cheapest model
GPT-4o	$2.50	$10.00	50% cut from 2024 ($5.00 input)
Claude Sonnet 4.6	$3.00	$15.00	Same price as Sonnet 4.5
Gemini 2.5 Pro	$1.25	$10.00	—
Claude Opus 4.6	$5.00	$25.00	—
Legacy Opus 4.1	$15.00	$75.00	3x more expensive than current Opus — check your config

What changed significantly:

• GPT-4o input price dropped from $5.00 to $2.50/1M tokens — a 50% reduction.
• Mistral Large 3 dropped from $8/$24 to $2/$6 per million tokens — a 75% cut on input.
• Perplexity Sonar: citation tokens no longer billed as of early 2026. Reduces per-response cost for high-volume search augmentation.
• Google Gemini context caching now available — reduces costs up to 75% on long-context tasks.

Free tiers on LLM APIs:

Google Gemini is the only major frontier model with a genuinely usable free tier: 1.5M tokens/day via AI Studio. Mistral's free tier via La Plateforme is rate-limited but sufficient for testing.

---

AI Coding Assistants

Tool	Free Tier	Paid Tier	Notes
GitHub Copilot	2,000 completions + 50 chats/mo	$10/mo (Pro)	GPT-4o and Claude Sonnet on free tier
Cursor	2,000 completions + 50 slow requests/mo	$20/mo (Pro)	Pro+ $60/mo; Ultra $200/mo
Amazon Q Developer	50 agentic req + unlimited inline/mo	$19/user/mo (Pro)	Formerly CodeWhisperer
Codeium (Windsurf)	Unlimited completions	$15/mo (Pro)	Rebranded to Windsurf in 2025

GitHub Copilot's late-2024 free tier launch was significant — it provides GPT-4o and Claude Sonnet access at zero cost. The 50 chat messages/month limit is the binding constraint for most users.

Cursor's tier structure expanded in 2025: Pro ($20/mo), Pro+ ($60/mo with higher request limits), and Ultra ($200/mo for power users). Business plans at $40/user/mo.

---

Vector Databases

Tool	Free Tier	Starter Price	Self-Host Option
Pinecone	2GB, 5 indexes, 2M write units/mo	PAYG (no monthly minimum)	No
Qdrant Cloud	1GB RAM cluster (permanent, not trial)	~$45/mo (Starter)	Yes — open source, free
Weaviate	14-day sandbox only	$45/mo (Flex)	Yes — BSD-3 license
Chroma	No cloud free tier	Beta pricing	Yes — MIT license
Milvus/Zilliz	Serverless free tier	From $0 (serverless)	Yes — open source

Self-hosting math: Qdrant self-hosted on a $20/mo VPS costs $20/mo vs. $45/mo for Qdrant Cloud. The breakeven depends on your engineering time cost, not the raw compute cost. Weaviate's 14-day sandbox deadline is easy to miss — no credit card at signup, but you'll be asked before day 14 ends.

---

ML Platforms and Experiment Tracking

Tool	Free Tier	Paid Tier	Notes
Weights & Biases	Unlimited experiments, 100GB storage	$35/user/mo (Team)	No expiry; no seat limit on free
Comet ML	1 project, limited compute	$179/mo (Team)	—
MLflow	Free (open source)	Self-hosted only	No managed cloud free tier
Neptune.ai	200 hours of runs free	$29/mo (Individual)	—

Weights & Biases' free tier is one of the most generous in ML tooling — unlimited experiment tracking with 100GB storage and no time limit. 90%+ of teams use it on the free tier.

---

Speech and Audio

Tool	Free Tier	Starter Price	Notes
ElevenLabs	10,000 characters/mo (~10 min TTS)	$5/mo (Starter)	Expanded to Conversational AI in 2025
Deepgram	$200 credit on signup	$0.0077/min (Nova-3 PAYG)	Streaming costs 79% more than batch
OpenAI Whisper	Free (open source)	API: $0.006/min	Batch is cheapest; API adds latency
AssemblyAI	100 hours free	$0.007/min (Nano)	Strongest speaker diarization

Deepgram's streaming vs. batch cost difference matters at scale: streaming (real-time transcription) costs 79% more per minute than batch (async) on the Nova-3 model. If your use case allows async processing, the cost difference is significant.

---

AI Observability

Tool	Free Tier	Paid Tier	Notes
Langfuse	1M trace spans/mo, unlimited users	$29/mo (Hobby Cloud)	Self-hosted version: fully free, no limits
Helicone	10K requests/mo	$79/mo	Proxy-based — zero code change to instrument
Arize AI	None	$50,000/yr minimum	Enterprise-only
WhyLabs	None	$799/mo minimum	Enterprise-only

Langfuse is MIT licensed — the self-hosted version is completely free with no limits on traces, users, or retention. For high-volume production LLM apps, self-hosted Langfuse avoids the $29+/month cloud cost.

---

What's in the Full Dataset

57 tools, 10 categories, 15 columns per row including:

• Exact free tier limits with no vague descriptions ("generous free tier" is not useful — we use specific numbers)
• Starter and mid-tier pricing, monthly and annual
• Enterprise pricing notes and ranges from Vendr/G2 buyer data where available
• Self-hosting availability and license type
• 2025–2026 pricing changes, rebrands, and acquisitions for every tool
• Direct pricing URLs, all verified March 2026

---

All pricing verified March 2026 from public pricing pages and published API documentation. Token prices in particular change frequently — verify before budgeting.